from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import logging
import functools
import numpy as np
import torch
import torch.nn as nn
import torch._utils
import torch.nn.functional as F
import sys
sys.path.append('/home/ubuntu/ds/segmentation/HRNet-Semantic-Segmentation/lib/models/')
# from .bn_helper import BatchNorm2d, BatchNorm2d_class, relu_inplace
from bn_helper import BatchNorm2d, BatchNorm2d_class, relu_inplace
= True
ALIGN_CORNERS = 0.1
BN_MOMENTUM = logging.getLogger(__name__)
logger '/home/ubuntu/ds/segmentation/HRNet-Semantic-Segmentation/lib/')
sys.path.append(from config import config
from config import update_config
from config.default import _C as cfg
# config file
= '/home/ubuntu/ds/segmentation/HRNet-Semantic-Segmentation/experiments/cityscapes/seg_hrnet_ocr_w48_trainval_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484.yaml'
path
config.merge_from_file(path)
class ModuleHelper:
@staticmethod
def BNReLU(num_features, bn_type=None, **kwargs):
return nn.Sequential(
**kwargs),
BatchNorm2d(num_features,
nn.ReLU()
)
@staticmethod
def BatchNorm2d(*args, **kwargs):
return BatchNorm2d
def conv3x3(in_planes, out_planes, stride=1):
"""3x3 convolution with padding"""
return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
=1, bias=False) padding
Hrnet + ocr module is as follows, all the codes borrow from :
1. https://github.com/HRNet/HRNet-Semantic-Segmentation/blob/HRNet-OCR/lib/models/seg_hrnet_ocr.py
2. https://github.com/openseg-group/openseg.pytorch
base part
config
CfgNode({'OUTPUT_DIR': 'output', 'LOG_DIR': 'log', 'GPUS': (0,), 'WORKERS': 1, 'PRINT_FREQ': 10, 'AUTO_RESUME': False, 'PIN_MEMORY': True, 'RANK': 0, 'CUDNN': CfgNode({'BENCHMARK': True, 'DETERMINISTIC': False, 'ENABLED': True}), 'MODEL': CfgNode({'NAME': 'seg_hrnet_ocr', 'PRETRAINED': 'pretrained_models/hrnetv2_w48_imagenet_pretrained.pth', 'ALIGN_CORNERS': True, 'NUM_OUTPUTS': 2, 'EXTRA': CfgNode({'FINAL_CONV_KERNEL': 1, 'STAGE1': CfgNode({'NUM_MODULES': 1, 'NUM_RANCHES': 1, 'BLOCK': 'BOTTLENECK', 'NUM_BLOCKS': [4], 'NUM_CHANNELS': [64], 'FUSE_METHOD': 'SUM'}), 'STAGE2': CfgNode({'NUM_MODULES': 1, 'NUM_BRANCHES': 2, 'BLOCK': 'BASIC', 'NUM_BLOCKS': [4, 4], 'NUM_CHANNELS': [48, 96], 'FUSE_METHOD': 'SUM'}), 'STAGE3': CfgNode({'NUM_MODULES': 4, 'NUM_BRANCHES': 3, 'BLOCK': 'BASIC', 'NUM_BLOCKS': [4, 4, 4], 'NUM_CHANNELS': [48, 96, 192], 'FUSE_METHOD': 'SUM'}), 'STAGE4': CfgNode({'NUM_MODULES': 3, 'NUM_BRANCHES': 4, 'BLOCK': 'BASIC', 'NUM_BLOCKS': [4, 4, 4, 4], 'NUM_CHANNELS': [48, 96, 192, 384], 'FUSE_METHOD': 'SUM'})}), 'OCR': CfgNode({'MID_CHANNELS': 512, 'KEY_CHANNELS': 256, 'DROPOUT': 0.05, 'SCALE': 1})}), 'LOSS': CfgNode({'USE_OHEM': False, 'OHEMTHRES': 0.9, 'OHEMKEEP': 131072, 'CLASS_BALANCE': False, 'BALANCE_WEIGHTS': [0.4, 1]}), 'DATASET': CfgNode({'ROOT': '', 'DATASET': 'cityscapesEXTRA_TRAIN_SET', 'NUM_CLASSES': 19, 'TRAIN_SET': 'list/cityscapes/trainval.lst', 'EXTRA_TRAIN_SET': '', 'TEST_SET': 'list/cityscapes/val.lst'}), 'TRAIN': CfgNode({'FREEZE_LAYERS': '', 'FREEZE_EPOCHS': -1, 'NONBACKBONE_KEYWORDS': [], 'NONBACKBONE_MULT': 10, 'IMAGE_SIZE': [1024, 512], 'BASE_SIZE': 2048, 'DOWNSAMPLERATE': 1, 'FLIP': True, 'MULTI_SCALE': True, 'SCALE_FACTOR': 16, 'RANDOM_BRIGHTNESS': False, 'RANDOM_BRIGHTNESS_SHIFT_VALUE': 10, 'LR_FACTOR': 0.1, 'LR_STEP': [90, 110], 'LR': 0.01, 'EXTRA_LR': 0.001, 'OPTIMIZER': 'sgd', 'MOMENTUM': 0.9, 'WD': 0.0005, 'NESTEROV': False, 'IGNORE_LABEL': 255, 'BEGIN_EPOCH': 0, 'END_EPOCH': 484, 'EXTRA_EPOCH': 0, 'RESUME': True, 'BATCH_SIZE_PER_GPU': 3, 'SHUFFLE': True, 'NUM_SAMPLES': 0}), 'TEST': CfgNode({'IMAGE_SIZE': [2048, 1024], 'BASE_SIZE': 2048, 'BATCH_SIZE_PER_GPU': 4, 'NUM_SAMPLES': 0, 'MODEL_FILE': '', 'FLIP_TEST': False, 'MULTI_SCALE': False, 'SCALE_LIST': [1], 'OUTPUT_INDEX': -1}), 'DEBUG': CfgNode({'DEBUG': False, 'SAVE_BATCH_IMAGES_GT': False, 'SAVE_BATCH_IMAGES_PRED': False, 'SAVE_HEATMAPS_GT': False, 'SAVE_HEATMAPS_PRED': False})})
for i in range(1,4):
print(i)
1
2
3
OCR module
class SpatialGather_Module(nn.Module):
"""
Aggregate the context features according to the initial
predicted probability distribution.
Employ the soft-weighted method to aggregate the context.
"""
def __init__(self, cls_num=0, scale=1):
super(SpatialGather_Module, self).__init__()
self.cls_num = cls_num
self.scale = scale
def forward(self, feats, probs):
= probs.size(0), probs.size(
batch_size, c, h, w 1), probs.size(2), probs.size(3)
= probs.view(batch_size, c, -1)
probs = feats.view(batch_size, feats.size(1), -1)
feats = feats.permute(0, 2, 1) # batch x hw x c
feats = F.softmax(self.scale * probs, dim=2) # batch x k x hw
probs = torch.matmul(probs, feats)\
ocr_context 0, 2, 1).unsqueeze(3) # batch x k x c
.permute(return ocr_context
class _ObjectAttentionBlock(nn.Module):
'''
The basic implementation for object context block
Input:
N X C X H X W
Parameters:
in_channels : the dimension of the input feature map
key_channels : the dimension after the key/query transform
scale : choose the scale to downsample the input feature maps (save memory cost)
bn_type : specify the bn type
Return:
N X C X H X W
'''
def __init__(self,
in_channels,
key_channels,=1,
scale=None):
bn_typesuper(_ObjectAttentionBlock, self).__init__()
self.scale = scale
self.in_channels = in_channels
self.key_channels = key_channels
self.pool = nn.MaxPool2d(kernel_size=(scale, scale))
self.f_pixel = nn.Sequential(
=self.in_channels, out_channels=self.key_channels,
nn.Conv2d(in_channels=1, stride=1, padding=0, bias=False),
kernel_sizeself.key_channels, bn_type=bn_type),
ModuleHelper.BNReLU(=self.key_channels, out_channels=self.key_channels,
nn.Conv2d(in_channels=1, stride=1, padding=0, bias=False),
kernel_sizeself.key_channels, bn_type=bn_type),
ModuleHelper.BNReLU(
)self.f_object = nn.Sequential(
=self.in_channels, out_channels=self.key_channels,
nn.Conv2d(in_channels=1, stride=1, padding=0, bias=False),
kernel_sizeself.key_channels, bn_type=bn_type),
ModuleHelper.BNReLU(=self.key_channels, out_channels=self.key_channels,
nn.Conv2d(in_channels=1, stride=1, padding=0, bias=False),
kernel_sizeself.key_channels, bn_type=bn_type),
ModuleHelper.BNReLU(
)self.f_down = nn.Sequential(
=self.in_channels, out_channels=self.key_channels,
nn.Conv2d(in_channels=1, stride=1, padding=0, bias=False),
kernel_sizeself.key_channels, bn_type=bn_type),
ModuleHelper.BNReLU(
)self.f_up = nn.Sequential(
=self.key_channels, out_channels=self.in_channels,
nn.Conv2d(in_channels=1, stride=1, padding=0, bias=False),
kernel_sizeself.in_channels, bn_type=bn_type),
ModuleHelper.BNReLU(
)
def forward(self, x, proxy):
= x.size(0), x.size(2), x.size(3)
batch_size, h, w if self.scale > 1:
= self.pool(x)
x
= self.f_pixel(x).view(batch_size, self.key_channels, -1)
query = query.permute(0, 2, 1)
query = self.f_object(proxy).view(batch_size, self.key_channels, -1)
key = self.f_down(proxy).view(batch_size, self.key_channels, -1)
value = value.permute(0, 2, 1)
value
= torch.matmul(query, key)
sim_map = (self.key_channels**-.5) * sim_map
sim_map = F.softmax(sim_map, dim=-1)
sim_map
# add bg context ...
= torch.matmul(sim_map, value)
context = context.permute(0, 2, 1).contiguous()
context = context.view(batch_size, self.key_channels, *x.size()[2:])
context = self.f_up(context)
context if self.scale > 1:
= F.interpolate(input=context, size=(
context ='bilinear', align_corners=ALIGN_CORNERS)
h, w), mode
return context
class ObjectAttentionBlock2D(_ObjectAttentionBlock):
def __init__(self,
in_channels,
key_channels,=1,
scale=None):
bn_typesuper(ObjectAttentionBlock2D, self).__init__(in_channels,
key_channels,
scale,=bn_type)
bn_type
class SpatialOCR_Module(nn.Module):
"""
Implementation of the OCR module:
We aggregate the global object representation to update the representation for each pixel.
"""
def __init__(self,
in_channels,
key_channels,
out_channels,=1,
scale=0.1,
dropout=None):
bn_typesuper(SpatialOCR_Module, self).__init__()
self.object_context_block = ObjectAttentionBlock2D(in_channels,
key_channels,
scale,
bn_type)= 2 * in_channels
_in_channels
self.conv_bn_dropout = nn.Sequential(
nn.Conv2d(_in_channels, out_channels,=1, padding=0, bias=False),
kernel_size=bn_type),
ModuleHelper.BNReLU(out_channels, bn_type
nn.Dropout2d(dropout)
)
def forward(self, feats, proxy_feats):
= self.object_context_block(feats, proxy_feats)
context
= self.conv_bn_dropout(torch.cat([context, feats], 1))
output
return output
Basic and Bottleneck Module
class BasicBlock(nn.Module):
= 1
expansion
def __init__(self, inplanes, planes, stride=1, downsample=None):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn1 = BatchNorm2d(planes, momentum=BN_MOMENTUM)
self.relu = nn.ReLU(inplace=relu_inplace)
self.conv2 = conv3x3(planes, planes)
self.bn2 = BatchNorm2d(planes, momentum=BN_MOMENTUM)
self.downsample = downsample
self.stride = stride
def forward(self, x):
= x
residual
= self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out
= self.conv2(out)
out = self.bn2(out)
out
if self.downsample is not None:
= self.downsample(x)
residual
= out + residual
out = self.relu(out)
out
return out
class Bottleneck(nn.Module):
= 4
expansion
def __init__(self, inplanes, planes, stride=1, downsample=None):
super(Bottleneck, self).__init__()
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
self.bn1 = BatchNorm2d(planes, momentum=BN_MOMENTUM)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
=1, bias=False)
paddingself.bn2 = BatchNorm2d(planes, momentum=BN_MOMENTUM)
self.conv3 = nn.Conv2d(planes, planes * self.expansion, kernel_size=1,
=False)
biasself.bn3 = BatchNorm2d(planes * self.expansion,
=BN_MOMENTUM)
momentumself.relu = nn.ReLU(inplace=relu_inplace)
self.downsample = downsample
self.stride = stride
def forward(self, x):
= x
residual
= self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out
= self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out
= self.conv3(out)
out = self.bn3(out)
out
if self.downsample is not None:
= self.downsample(x)
residual
= out + residual
out = self.relu(out)
out
return out
HighResolution Module
class HighResolutionModule(nn.Module):
def __init__(self, num_branches, blocks, num_blocks, num_inchannels,
=True):
num_channels, fuse_method, multi_scale_outputsuper(HighResolutionModule, self).__init__()
self._check_branches(
num_branches, blocks, num_blocks, num_inchannels, num_channels)
self.num_inchannels = num_inchannels
self.fuse_method = fuse_method
self.num_branches = num_branches
self.multi_scale_output = multi_scale_output
self.branches = self._make_branches(
num_branches, blocks, num_blocks, num_channels)self.fuse_layers = self._make_fuse_layers()
self.relu = nn.ReLU(inplace=relu_inplace)
def _check_branches(self, num_branches, blocks, num_blocks,
num_inchannels, num_channels):if num_branches != len(num_blocks):
= 'NUM_BRANCHES({}) <> NUM_BLOCKS({})'.format(
error_msg len(num_blocks))
num_branches,
logger.error(error_msg)raise ValueError(error_msg)
if num_branches != len(num_channels):
= 'NUM_BRANCHES({}) <> NUM_CHANNELS({})'.format(
error_msg len(num_channels))
num_branches,
logger.error(error_msg)raise ValueError(error_msg)
if num_branches != len(num_inchannels):
= 'NUM_BRANCHES({}) <> NUM_INCHANNELS({})'.format(
error_msg len(num_inchannels))
num_branches,
logger.error(error_msg)raise ValueError(error_msg)
def _make_one_branch(self, branch_index, block, num_blocks, num_channels,
=1):
stride= None
downsample if stride != 1 or \
self.num_inchannels[branch_index] != num_channels[branch_index] * block.expansion:
= nn.Sequential(
downsample self.num_inchannels[branch_index],
nn.Conv2d(* block.expansion,
num_channels[branch_index] =1, stride=stride, bias=False),
kernel_size* block.expansion,
BatchNorm2d(num_channels[branch_index] =BN_MOMENTUM),
momentum
)
= []
layers self.num_inchannels[branch_index],
layers.append(block(
num_channels[branch_index], stride, downsample))self.num_inchannels[branch_index] = \
* block.expansion
num_channels[branch_index] for i in range(1, num_blocks[branch_index]):
self.num_inchannels[branch_index],
layers.append(block(
num_channels[branch_index]))
return nn.Sequential(*layers)
def _make_branches(self, num_branches, block, num_blocks, num_channels):
= []
branches
for i in range(num_branches):
branches.append(self._make_one_branch(i, block, num_blocks, num_channels))
return nn.ModuleList(branches)
def _make_fuse_layers(self):
if self.num_branches == 1:
return None
= self.num_branches
num_branches = self.num_inchannels
num_inchannels = []
fuse_layers for i in range(num_branches if self.multi_scale_output else 1):
= []
fuse_layer for j in range(num_branches):
if j > i:
fuse_layer.append(nn.Sequential(
nn.Conv2d(num_inchannels[j],
num_inchannels[i],1,
1,
0,
=False),
bias=BN_MOMENTUM)))
BatchNorm2d(num_inchannels[i], momentumelif j == i:
None)
fuse_layer.append(else:
= []
conv3x3s for k in range(i-j):
if k == i - j - 1:
= num_inchannels[i]
num_outchannels_conv3x3
conv3x3s.append(nn.Sequential(
nn.Conv2d(num_inchannels[j],
num_outchannels_conv3x3,3, 2, 1, bias=False),
BatchNorm2d(num_outchannels_conv3x3,=BN_MOMENTUM)))
momentumelse:
= num_inchannels[j]
num_outchannels_conv3x3
conv3x3s.append(nn.Sequential(
nn.Conv2d(num_inchannels[j],
num_outchannels_conv3x3,3, 2, 1, bias=False),
BatchNorm2d(num_outchannels_conv3x3,=BN_MOMENTUM),
momentum=relu_inplace)))
nn.ReLU(inplace*conv3x3s))
fuse_layer.append(nn.Sequential(
fuse_layers.append(nn.ModuleList(fuse_layer))
return nn.ModuleList(fuse_layers)
def get_num_inchannels(self):
return self.num_inchannels
def forward(self, x):
if self.num_branches == 1:
return [self.branches[0](x[0])]
for i in range(self.num_branches):
= self.branches[i](x[i])
x[i]
= []
x_fuse for i in range(len(self.fuse_layers)):
= x[0] if i == 0 else self.fuse_layers[i][0](x[0])
y for j in range(1, self.num_branches):
if i == j:
= y + x[j]
y elif j > i:
= x[i].shape[-1]
width_output = x[i].shape[-2]
height_output = y + F.interpolate(
y self.fuse_layers[i][j](x[j]),
=[height_output, width_output],
size='bilinear', align_corners=ALIGN_CORNERS)
modeelse:
= y + self.fuse_layers[i][j](x[j])
y self.relu(y))
x_fuse.append(
return x_fuse
= {
blocks_dict 'BASIC': BasicBlock,
'BOTTLENECK': Bottleneck
}
class HighResolutionNet(nn.Module):
def __init__(self, config, **kwargs):
global ALIGN_CORNERS
= config.MODEL.EXTRA
extra super(HighResolutionNet, self).__init__()
= config.MODEL.ALIGN_CORNERS
ALIGN_CORNERS
# stem net
self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=2, padding=1,
=False)
biasself.bn1 = BatchNorm2d(64, momentum=BN_MOMENTUM)
self.conv2 = nn.Conv2d(64, 64, kernel_size=3, stride=2, padding=1,
=False)
biasself.bn2 = BatchNorm2d(64, momentum=BN_MOMENTUM)
self.relu = nn.ReLU(inplace=relu_inplace)
self.stage1_cfg = extra['STAGE1']
= self.stage1_cfg['NUM_CHANNELS'][0]
num_channels = blocks_dict[self.stage1_cfg['BLOCK']]
block = self.stage1_cfg['NUM_BLOCKS'][0]
num_blocks self.layer1 = self._make_layer(block, 64, num_channels, num_blocks)
= block.expansion*num_channels
stage1_out_channel
self.stage2_cfg = extra['STAGE2']
= self.stage2_cfg['NUM_CHANNELS']
num_channels = blocks_dict[self.stage2_cfg['BLOCK']]
block = [
num_channels * block.expansion for i in range(len(num_channels))]
num_channels[i] self.transition1 = self._make_transition_layer(
[stage1_out_channel], num_channels)self.stage2, pre_stage_channels = self._make_stage(
self.stage2_cfg, num_channels)
self.stage3_cfg = extra['STAGE3']
= self.stage3_cfg['NUM_CHANNELS']
num_channels = blocks_dict[self.stage3_cfg['BLOCK']]
block = [
num_channels * block.expansion for i in range(len(num_channels))]
num_channels[i] self.transition2 = self._make_transition_layer(
pre_stage_channels, num_channels)self.stage3, pre_stage_channels = self._make_stage(
self.stage3_cfg, num_channels)
self.stage4_cfg = extra['STAGE4']
= self.stage4_cfg['NUM_CHANNELS']
num_channels = blocks_dict[self.stage4_cfg['BLOCK']]
block = [
num_channels * block.expansion for i in range(len(num_channels))]
num_channels[i] self.transition3 = self._make_transition_layer(
pre_stage_channels, num_channels)self.stage4, pre_stage_channels = self._make_stage(
self.stage4_cfg, num_channels, multi_scale_output=True)
= np.int(np.sum(pre_stage_channels))
last_inp_channels = config.MODEL.OCR.MID_CHANNELS
ocr_mid_channels = config.MODEL.OCR.KEY_CHANNELS
ocr_key_channels
self.conv3x3_ocr = nn.Sequential(
nn.Conv2d(last_inp_channels, ocr_mid_channels,=3, stride=1, padding=1),
kernel_size
BatchNorm2d(ocr_mid_channels),=relu_inplace),
nn.ReLU(inplace
)self.ocr_gather_head = SpatialGather_Module(config.DATASET.NUM_CLASSES)
self.ocr_distri_head = SpatialOCR_Module(in_channels=ocr_mid_channels,
=ocr_key_channels,
key_channels=ocr_mid_channels,
out_channels=1,
scale=0.05,
dropout
)self.cls_head = nn.Conv2d(
=1, stride=1, padding=0, bias=True)
ocr_mid_channels, config.DATASET.NUM_CLASSES, kernel_size
self.aux_head = nn.Sequential(
nn.Conv2d(last_inp_channels, last_inp_channels,=1, stride=1, padding=0),
kernel_size
BatchNorm2d(last_inp_channels),=relu_inplace),
nn.ReLU(inplace
nn.Conv2d(last_inp_channels, config.DATASET.NUM_CLASSES,=1, stride=1, padding=0, bias=True)
kernel_size
)
def _make_transition_layer(
self, num_channels_pre_layer, num_channels_cur_layer):
= len(num_channels_cur_layer)
num_branches_cur = len(num_channels_pre_layer)
num_branches_pre = []
transition_layers for i in range(num_branches_cur):
if i < num_branches_pre:
if num_channels_cur_layer[i] != num_channels_pre_layer[i]:
transition_layers.append(nn.Sequential(
nn.Conv2d(num_channels_pre_layer[i],
num_channels_cur_layer[i],3,
1,
1,
=False),
bias
BatchNorm2d(=BN_MOMENTUM),
num_channels_cur_layer[i], momentum=relu_inplace)))
nn.ReLU(inplaceelse:
None)
transition_layers.append(else:
= []
conv3x3s for j in range(i+1-num_branches_pre):
= num_channels_pre_layer[-1]
inchannels = num_channels_cur_layer[i] \
outchannels if j == i-num_branches_pre else inchannels
conv3x3s.append(nn.Sequential(
nn.Conv2d(3, 2, 1, bias=False),
inchannels, outchannels, =BN_MOMENTUM),
BatchNorm2d(outchannels, momentum=relu_inplace)))
nn.ReLU(inplace*conv3x3s))
transition_layers.append(nn.Sequential(
return nn.ModuleList(transition_layers)
def _make_layer(self, block, inplanes, planes, blocks, stride=1):
= None
downsample if stride != 1 or inplanes != planes * block.expansion:
= nn.Sequential(
downsample * block.expansion,
nn.Conv2d(inplanes, planes =1, stride=stride, bias=False),
kernel_size* block.expansion, momentum=BN_MOMENTUM),
BatchNorm2d(planes
)
= []
layers
layers.append(block(inplanes, planes, stride, downsample))= planes * block.expansion
inplanes for i in range(1, blocks):
layers.append(block(inplanes, planes))
return nn.Sequential(*layers)
def _make_stage(self, layer_config, num_inchannels,
=True):
multi_scale_output= layer_config['NUM_MODULES']
num_modules = layer_config['NUM_BRANCHES']
num_branches = layer_config['NUM_BLOCKS']
num_blocks = layer_config['NUM_CHANNELS']
num_channels = blocks_dict[layer_config['BLOCK']]
block = layer_config['FUSE_METHOD']
fuse_method
= []
modules for i in range(num_modules):
# multi_scale_output is only used last module
if not multi_scale_output and i == num_modules - 1:
= False
reset_multi_scale_output else:
= True
reset_multi_scale_output
modules.append(
HighResolutionModule(num_branches,
block,
num_blocks,
num_inchannels,
num_channels,
fuse_method,
reset_multi_scale_output)
)= modules[-1].get_num_inchannels()
num_inchannels
return nn.Sequential(*modules), num_inchannels
def forward(self, x):
= self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.conv2(x)
x = self.bn2(x)
x = self.relu(x)
x = self.layer1(x)
x
= []
x_list for i in range(self.stage2_cfg['NUM_BRANCHES']):
if self.transition1[i] is not None:
self.transition1[i](x))
x_list.append(else:
x_list.append(x)= self.stage2(x_list)
y_list
= []
x_list for i in range(self.stage3_cfg['NUM_BRANCHES']):
if self.transition2[i] is not None:
if i < self.stage2_cfg['NUM_BRANCHES']:
self.transition2[i](y_list[i]))
x_list.append(else:
self.transition2[i](y_list[-1]))
x_list.append(else:
x_list.append(y_list[i])= self.stage3(x_list)
y_list
= []
x_list for i in range(self.stage4_cfg['NUM_BRANCHES']):
if self.transition3[i] is not None:
if i < self.stage3_cfg['NUM_BRANCHES']:
self.transition3[i](y_list[i]))
x_list.append(else:
self.transition3[i](y_list[-1]))
x_list.append(else:
x_list.append(y_list[i])= self.stage4(x_list)
x
# Upsampling
= x[0].size(2), x[0].size(3)
x0_h, x0_w = F.interpolate(x[1], size=(x0_h, x0_w),
x1 ='bilinear', align_corners=ALIGN_CORNERS)
mode= F.interpolate(x[2], size=(x0_h, x0_w),
x2 ='bilinear', align_corners=ALIGN_CORNERS)
mode= F.interpolate(x[3], size=(x0_h, x0_w),
x3 ='bilinear', align_corners=ALIGN_CORNERS)
mode
= torch.cat([x[0], x1, x2, x3], 1)
feats
= []
out_aux_seg
# ocr
= self.aux_head(feats)
out_aux # compute contrast feature
= self.conv3x3_ocr(feats)
feats
= self.ocr_gather_head(feats, out_aux)
context = self.ocr_distri_head(feats, context)
feats
= self.cls_head(feats)
out
out_aux_seg.append(out_aux)
out_aux_seg.append(out)
return out_aux_seg
def init_weights(self, pretrained='',):
'=> init weights from normal distribution')
logger.info(for name, m in self.named_modules():
if any(part in name for part in {'cls', 'aux', 'ocr'}):
# print('skipped', name)
continue
if isinstance(m, nn.Conv2d):
=0.001)
nn.init.normal_(m.weight, stdelif isinstance(m, BatchNorm2d_class):
1)
nn.init.constant_(m.weight, 0)
nn.init.constant_(m.bias, if os.path.isfile(pretrained):
= torch.load(
pretrained_dict ={'cuda:0': 'cpu'})
pretrained, map_location'=> loading pretrained model {}'.format(pretrained))
logger.info(= self.state_dict()
model_dict = {k.replace('last_layer', 'aux_head').replace(
pretrained_dict 'model.', ''): v for k, v in pretrained_dict.items()}
print(set(model_dict) - set(pretrained_dict))
print(set(pretrained_dict) - set(model_dict))
= {k: v for k, v in pretrained_dict.items()
pretrained_dict if k in model_dict.keys()}
# for k, _ in pretrained_dict.items():
# logger.info(
# '=> loading {} pretrained model {}'.format(k, pretrained))
model_dict.update(pretrained_dict)self.load_state_dict(model_dict)
elif pretrained:
raise RuntimeError('No such file {}'.format(pretrained))
relu
fuse_layer
Test model
def get_seg_model(cfg, **kwargs):
= HighResolutionNet(cfg, **kwargs)
model # model.init_weights(cfg.MODEL.PRETRAINED)
return model
= get_seg_model(config)
model
model.cuda()= torch.randn(1, 3, 256, 256).cuda()
x = model(x)
out print(out)
HighResolutionNet(
(conv1): Conv2d(3, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(layer1): Sequential(
(0): Bottleneck(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(transition1): ModuleList(
(0): Sequential(
(0): Conv2d(256, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(256, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
)
)
(stage2): Sequential(
(0): HighResolutionModule(
(branches): ModuleList(
(0): Sequential(
(0): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(fuse_layers): ModuleList(
(0): ModuleList(
(0): None
(1): Sequential(
(0): Conv2d(96, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): None
)
)
(relu): ReLU(inplace=True)
)
)
(transition2): ModuleList(
(0): None
(1): None
(2): Sequential(
(0): Sequential(
(0): Conv2d(96, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
)
)
(stage3): Sequential(
(0): HighResolutionModule(
(branches): ModuleList(
(0): Sequential(
(0): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(fuse_layers): ModuleList(
(0): ModuleList(
(0): None
(1): Sequential(
(0): Conv2d(96, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): Sequential(
(0): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): None
(2): Sequential(
(0): Conv2d(192, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): None
)
)
(relu): ReLU(inplace=True)
)
(1): HighResolutionModule(
(branches): ModuleList(
(0): Sequential(
(0): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(fuse_layers): ModuleList(
(0): ModuleList(
(0): None
(1): Sequential(
(0): Conv2d(96, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): Sequential(
(0): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): None
(2): Sequential(
(0): Conv2d(192, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): None
)
)
(relu): ReLU(inplace=True)
)
(2): HighResolutionModule(
(branches): ModuleList(
(0): Sequential(
(0): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(fuse_layers): ModuleList(
(0): ModuleList(
(0): None
(1): Sequential(
(0): Conv2d(96, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): Sequential(
(0): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): None
(2): Sequential(
(0): Conv2d(192, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): None
)
)
(relu): ReLU(inplace=True)
)
(3): HighResolutionModule(
(branches): ModuleList(
(0): Sequential(
(0): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(fuse_layers): ModuleList(
(0): ModuleList(
(0): None
(1): Sequential(
(0): Conv2d(96, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): Sequential(
(0): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): None
(2): Sequential(
(0): Conv2d(192, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): None
)
)
(relu): ReLU(inplace=True)
)
)
(transition3): ModuleList(
(0): None
(1): None
(2): None
(3): Sequential(
(0): Sequential(
(0): Conv2d(192, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
)
)
(stage4): Sequential(
(0): HighResolutionModule(
(branches): ModuleList(
(0): Sequential(
(0): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): Sequential(
(0): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(fuse_layers): ModuleList(
(0): ModuleList(
(0): None
(1): Sequential(
(0): Conv2d(96, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): Sequential(
(0): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): Sequential(
(0): Conv2d(384, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): None
(2): Sequential(
(0): Conv2d(192, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): Sequential(
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): None
(3): Sequential(
(0): Conv2d(384, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(2): Sequential(
(0): Conv2d(48, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): Sequential(
(0): Conv2d(192, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): None
)
)
(relu): ReLU(inplace=True)
)
(1): HighResolutionModule(
(branches): ModuleList(
(0): Sequential(
(0): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): Sequential(
(0): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(fuse_layers): ModuleList(
(0): ModuleList(
(0): None
(1): Sequential(
(0): Conv2d(96, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): Sequential(
(0): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): Sequential(
(0): Conv2d(384, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): None
(2): Sequential(
(0): Conv2d(192, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): Sequential(
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): None
(3): Sequential(
(0): Conv2d(384, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(2): Sequential(
(0): Conv2d(48, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): Sequential(
(0): Conv2d(192, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): None
)
)
(relu): ReLU(inplace=True)
)
(2): HighResolutionModule(
(branches): ModuleList(
(0): Sequential(
(0): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): Sequential(
(0): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): BasicBlock(
(conv1): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(fuse_layers): ModuleList(
(0): ModuleList(
(0): None
(1): Sequential(
(0): Conv2d(96, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(2): Sequential(
(0): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): Sequential(
(0): Conv2d(384, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): None
(2): Sequential(
(0): Conv2d(192, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(3): Sequential(
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): None
(3): Sequential(
(0): Conv2d(384, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): ModuleList(
(0): Sequential(
(0): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(48, 48, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(2): Sequential(
(0): Conv2d(48, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Sequential(
(0): Sequential(
(0): Conv2d(96, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(1): Sequential(
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(2): Sequential(
(0): Sequential(
(0): Conv2d(192, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(3): None
)
)
(relu): ReLU(inplace=True)
)
)
(conv3x3_ocr): Sequential(
(0): Conv2d(720, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(ocr_gather_head): SpatialGather_Module()
(ocr_distri_head): SpatialOCR_Module(
(object_context_block): ObjectAttentionBlock2D(
(pool): MaxPool2d(kernel_size=(1, 1), stride=(1, 1), padding=0, dilation=1, ceil_mode=False)
(f_pixel): Sequential(
(0): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): Sequential(
(0): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): ReLU()
)
(2): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(3): Sequential(
(0): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): ReLU()
)
)
(f_object): Sequential(
(0): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): Sequential(
(0): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): ReLU()
)
(2): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(3): Sequential(
(0): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): ReLU()
)
)
(f_down): Sequential(
(0): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): Sequential(
(0): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): ReLU()
)
)
(f_up): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): Sequential(
(0): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): ReLU()
)
)
)
(conv_bn_dropout): Sequential(
(0): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): Sequential(
(0): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): ReLU()
)
(2): Dropout2d(p=0.05, inplace=False)
)
)
(cls_head): Conv2d(512, 19, kernel_size=(1, 1), stride=(1, 1))
(aux_head): Sequential(
(0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(720, 19, kernel_size=(1, 1), stride=(1, 1))
)
)
[tensor([[[[ 0.0126, 0.2089, 0.1555, ..., -0.2633, -0.2207, -0.1966],
[-0.0301, 0.0580, 0.3048, ..., -0.0795, -0.3500, -0.5949],
[-0.0560, 0.4462, 0.6888, ..., -0.0914, -0.1904, -0.5955],
...,
[-0.2828, -0.3316, -0.3327, ..., 0.0599, -0.1506, 0.1004],
[-0.4286, -0.1721, -0.4987, ..., 0.1643, -0.1669, -0.2276],
[-0.4279, -0.4016, -0.3705, ..., -0.2004, -0.3496, -0.6257]],
[[ 0.2175, 0.1642, 0.2870, ..., -0.2048, -0.1047, -0.1548],
[ 0.1737, 0.1060, 0.1426, ..., -0.0630, -0.2284, -0.1860],
[ 0.1276, -0.0596, -0.0157, ..., -0.3222, -0.4208, -0.2417],
...,
[ 0.2149, -0.1225, 0.2284, ..., -0.0518, -0.3060, -0.1962],
[-0.1730, -0.2476, 0.0653, ..., -0.0104, -0.0105, -0.0807],
[-0.3364, -0.3119, -0.2409, ..., -0.0378, 0.0649, 0.0647]],
[[-0.5266, -0.3146, -0.0793, ..., -0.0764, 0.0503, 0.1208],
[-0.6032, -0.5813, -0.3442, ..., -0.4756, -0.2503, -0.1254],
[-0.2866, -0.6633, -0.3280, ..., -0.3490, -0.3198, 0.1013],
...,
[-0.2102, -0.1872, -0.3686, ..., 0.2620, 0.3569, 0.5164],
[-0.2391, -0.0143, -0.3167, ..., 0.1354, 0.4852, 0.4345],
[-0.0980, -0.0999, -0.2332, ..., -0.0388, 0.2441, 0.2310]],
...,
[[-0.1700, -0.0286, -0.0042, ..., -0.8255, -0.8084, -0.5332],
[ 0.0808, 0.1889, 0.0577, ..., -0.6874, -0.4808, -0.2507],
[ 0.1011, 0.1252, -0.0381, ..., -0.4298, -0.4084, -0.1023],
...,
[ 0.2143, 0.0511, 0.0697, ..., -0.8343, -0.7336, -0.4068],
[ 0.3050, -0.1730, -0.1157, ..., -0.8594, -0.7523, -0.6235],
[ 0.1338, -0.2147, -0.3771, ..., -0.7364, -0.7153, -0.7533]],
[[-0.6849, -0.3476, -0.0539, ..., -0.2995, -0.1954, -0.1797],
[-0.6074, -0.4295, -0.3755, ..., -0.4582, -0.4791, -0.5893],
[-0.5042, -0.6442, -0.4120, ..., -0.7364, -0.8572, -0.6877],
...,
[ 0.1772, 0.1758, 0.1408, ..., 0.2219, 0.3596, 0.1636],
[ 0.0977, 0.2228, 0.0980, ..., -0.2784, 0.1537, 0.0426],
[ 0.2082, 0.1112, 0.3127, ..., -0.1324, 0.2484, 0.0712]],
[[-0.7520, -0.7920, -0.8023, ..., -0.3344, -0.4756, -0.3786],
[-0.5891, -0.5307, -0.6568, ..., -0.4531, -0.4720, -0.4871],
[-0.7524, -0.7254, -0.7203, ..., -0.5278, -0.4140, -0.3738],
...,
[-0.3742, -0.3159, -0.2781, ..., -0.8083, -0.8444, -1.0273],
[-0.3527, -0.2352, -0.1589, ..., -0.6653, -0.9925, -1.2293],
[-0.4631, -0.5654, -0.5520, ..., -1.1085, -1.1969, -1.6035]]]],
device='cuda:0', grad_fn=<CudnnConvolutionBackward>), tensor([[[[-1.1921e+00, -8.7566e-01, -7.3220e-01, ..., -7.0549e-01,
-1.0782e+00, 1.6466e-01],
[-5.1973e-01, -3.2829e-01, -4.0623e-02, ..., -3.4792e-01,
-2.0699e-01, 6.5108e-02],
[-5.8162e-01, 4.2323e-03, 1.0330e-01, ..., -7.2836e-03,
5.7834e-02, -4.0695e-02],
...,
[-2.0400e-01, -1.0644e-01, -5.4366e-02, ..., -1.6851e-01,
-1.5228e-01, 4.7855e-01],
[-4.7569e-01, 9.6399e-02, -2.8926e-01, ..., -8.0138e-02,
-5.0628e-01, 3.7228e-01],
[ 9.3950e-02, -4.9347e-02, -1.0575e-01, ..., -9.6701e-01,
-1.0447e+00, -2.0369e-02]],
[[ 1.3028e+00, 6.7632e-01, 9.9330e-01, ..., 9.1765e-01,
8.9304e-01, 9.7930e-01],
[ 6.8199e-01, 4.9716e-01, 3.4814e-01, ..., 7.7114e-01,
6.6421e-01, 3.5449e-01],
[ 9.0915e-01, 1.0235e-01, 1.8720e-02, ..., 4.4257e-01,
3.4798e-01, 2.1372e-01],
...,
[ 9.0561e-01, -1.4279e-01, 5.0179e-02, ..., 1.8157e-01,
-1.3743e-01, 2.9420e-01],
[ 9.3803e-01, -1.0436e-02, 1.4767e-01, ..., 4.9976e-02,
1.0739e-01, 2.6171e-01],
[ 9.2503e-01, 3.5870e-01, 4.8831e-01, ..., 8.4631e-01,
3.8659e-01, 3.3774e-01]],
[[ 2.0981e-01, 6.1833e-01, 9.6794e-01, ..., 2.8238e-01,
2.5978e-01, 4.8613e-01],
[-1.2846e-01, 6.7296e-02, -6.3639e-02, ..., -4.2966e-01,
3.1935e-01, 5.5187e-01],
[ 2.5797e-01, -1.2030e-01, -2.0252e-01, ..., 1.9528e-01,
1.5962e-01, 3.3269e-01],
...,
[-2.0601e-02, 8.2864e-02, -3.6000e-01, ..., 3.5094e-01,
-5.5201e-02, -2.5831e-02],
[-3.8858e-01, -4.6351e-02, -2.8292e-01, ..., 3.5836e-02,
-2.4179e-01, -1.3999e-03],
[-8.9333e-02, 4.3436e-01, 4.7767e-01, ..., 9.1226e-01,
7.1973e-01, 5.8763e-01]],
...,
[[ 8.5679e-01, 1.1779e-01, 3.7987e-01, ..., 3.8945e-01,
5.4756e-01, 5.8113e-01],
[ 2.1517e-01, 2.9140e-01, 8.3521e-02, ..., 4.1642e-01,
4.8541e-02, 5.7063e-01],
[ 4.2755e-01, 8.5772e-02, 4.9619e-02, ..., 1.8989e-01,
1.8816e-03, 4.7896e-01],
...,
[ 3.2219e-02, -3.5875e-02, 2.6908e-01, ..., 2.4972e-01,
2.1703e-01, 8.5617e-01],
[ 3.2784e-01, -6.2063e-02, 1.2819e-01, ..., 1.4086e-01,
2.0888e-01, 1.0915e+00],
[ 2.8844e-01, 5.4034e-02, 1.2540e-02, ..., 1.4383e+00,
1.2895e+00, 1.3019e+00]],
[[ 3.3014e-01, 2.7361e-01, -6.8319e-01, ..., 1.3882e-01,
9.3643e-02, 1.0667e-02],
[ 2.2844e-01, -1.8923e-01, -3.9863e-01, ..., -5.1912e-01,
-7.6968e-01, -4.6036e-01],
[ 6.4837e-02, -1.6802e-01, -1.0226e-01, ..., -7.3331e-01,
-7.6095e-01, -6.2935e-01],
...,
[ 2.4629e-01, -3.2840e-01, 6.8417e-02, ..., -2.9780e-01,
-2.2339e-01, -6.2761e-01],
[-2.2035e-01, -2.5319e-01, 7.7725e-03, ..., -1.8956e-01,
-4.1416e-01, -6.3720e-01],
[-3.2191e-01, -2.5960e-01, -4.0099e-01, ..., -5.5858e-01,
-6.1042e-01, -3.6042e-01]],
[[-2.0594e-01, 3.6297e-01, 7.4933e-01, ..., 6.1509e-01,
6.3861e-01, 9.7792e-02],
[ 7.0288e-01, 1.9002e-01, 2.3223e-02, ..., 1.9716e-01,
2.1533e-01, 1.1496e-01],
[ 5.7000e-01, 2.8020e-01, -5.3618e-02, ..., 1.0633e-01,
3.1564e-01, 2.8380e-01],
...,
[ 5.9550e-01, 1.5407e-01, 3.3551e-01, ..., -1.9185e-02,
1.3940e-01, -5.7018e-01],
[ 4.7579e-01, 2.6538e-01, 4.5185e-01, ..., 1.8412e-01,
2.5192e-01, -2.6754e-01],
[ 7.6439e-01, -1.3286e-01, -4.0669e-01, ..., -8.3105e-01,
-6.9897e-01, 2.7346e-01]]]], device='cuda:0',
grad_fn=<CudnnConvolutionBackward>)]