# ------------------------------------------------------------------------------ # Copyright (c) Microsoft # Licensed under the MIT License. # Written by Bin Xiao (Bin.Xiao@microsoft.com) # Modified by Rainbowsecret (yuyua@microsoft.com) # "High-Resolution Representations for Labeling Pixels and Regions" # ------------------------------------------------------------------------------ from __future__ import absolute_import from __future__ import division from __future__ import print_function import os import pdb import torch import torch.nn as nn import torch.nn.functional as F import numpy as np from lib.models.tools.module_helper import ModuleHelper from lib.utils.tools.logger import Logger as Log if torch.__version__.startswith('1'): relu_inplace = True else: relu_inplace = False def conv3x3(in_planes, out_planes, stride=1): """3x3 convolution with padding""" return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False) class BasicBlock(nn.Module): expansion = 1 def __init__(self, inplanes, planes, stride=1, downsample=None, bn_type=None, bn_momentum=0.1): super(BasicBlock, self).__init__() self.conv1 = conv3x3(inplanes, planes, stride) self.bn1 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(planes, momentum=bn_momentum) self.relu = nn.ReLU(inplace=False) self.relu_in = nn.ReLU(inplace=True) self.conv2 = conv3x3(planes, planes) self.bn2 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(planes, momentum=bn_momentum) self.downsample = downsample self.stride = stride def forward(self, x): residual = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) if self.downsample is not None: residual = self.downsample(x) out = out + residual out = self.relu_in(out) return out class Bottleneck(nn.Module): expansion = 4 def __init__(self, inplanes, planes, stride=1, downsample=None, bn_type=None, bn_momentum=0.1): super(Bottleneck, self).__init__() self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False) self.bn1 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(planes, momentum=bn_momentum) self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False) self.bn2 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(planes, momentum=bn_momentum) self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False) self.bn3 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(planes * 4, momentum=bn_momentum) self.relu = nn.ReLU(inplace=False) self.relu_in = nn.ReLU(inplace=True) self.downsample = downsample self.stride = stride def forward(self, x): residual = x out = 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) if self.downsample is not None: residual = self.downsample(x) out = out + residual out = self.relu_in(out) return out class HighResolutionModule(nn.Module): def __init__(self, num_branches, blocks, num_blocks, num_inchannels, num_channels, fuse_method, multi_scale_output=True, bn_type=None, bn_momentum=0.1): super(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, bn_type=bn_type, bn_momentum=bn_momentum) self.fuse_layers = self._make_fuse_layers(bn_type=bn_type, bn_momentum=bn_momentum) self.relu = nn.ReLU(inplace=False) def _check_branches(self, num_branches, blocks, num_blocks, num_inchannels, num_channels): if num_branches != len(num_blocks): error_msg = 'NUM_BRANCHES({}) <> NUM_BLOCKS({})'.format( num_branches, len(num_blocks)) Log.error(error_msg) raise ValueError(error_msg) if num_branches != len(num_channels): error_msg = 'NUM_BRANCHES({}) <> NUM_CHANNELS({})'.format( num_branches, len(num_channels)) Log.error(error_msg) raise ValueError(error_msg) if num_branches != len(num_inchannels): error_msg = 'NUM_BRANCHES({}) <> NUM_INCHANNELS({})'.format( num_branches, len(num_inchannels)) Log.error(error_msg) raise ValueError(error_msg) def _make_one_branch(self, branch_index, block, num_blocks, num_channels, stride=1, bn_type=None, bn_momentum=0.1): downsample = None if stride != 1 or \ self.num_inchannels[branch_index] != num_channels[branch_index] * block.expansion: downsample = nn.Sequential( nn.Conv2d( self.num_inchannels[branch_index], num_channels[branch_index] * block.expansion, kernel_size=1, stride=stride, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)( num_channels[branch_index] * block.expansion, momentum=bn_momentum ), ) layers = [] layers.append( block( self.num_inchannels[branch_index], num_channels[branch_index], stride, downsample, bn_type=bn_type, bn_momentum=bn_momentum ) ) self.num_inchannels[branch_index] = \ num_channels[branch_index] * block.expansion for i in range(1, num_blocks[branch_index]): layers.append( block( self.num_inchannels[branch_index], num_channels[branch_index], bn_type=bn_type, bn_momentum=bn_momentum ) ) return nn.Sequential(*layers) def _make_branches(self, num_branches, block, num_blocks, num_channels, bn_type, bn_momentum=0.1): branches = [] for i in range(num_branches): branches.append( self._make_one_branch(i, block, num_blocks, num_channels, bn_type=bn_type, bn_momentum=bn_momentum) ) return nn.ModuleList(branches) def _make_fuse_layers(self, bn_type, bn_momentum=0.1): if self.num_branches == 1: return None num_branches = self.num_branches num_inchannels = self.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, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(num_inchannels[i], momentum=bn_momentum), ) ) elif j == i: fuse_layer.append(None) else: conv3x3s = [] for k in range(i - j): if k == i - j - 1: num_outchannels_conv3x3 = num_inchannels[i] conv3x3s.append( nn.Sequential( nn.Conv2d( num_inchannels[j], num_outchannels_conv3x3, 3, 2, 1, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(num_outchannels_conv3x3, momentum=bn_momentum) ) ) else: num_outchannels_conv3x3 = num_inchannels[j] conv3x3s.append( nn.Sequential( nn.Conv2d( num_inchannels[j], num_outchannels_conv3x3, 3, 2, 1, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(num_outchannels_conv3x3, momentum=bn_momentum), nn.ReLU(inplace=False) ) ) fuse_layer.append(nn.Sequential(*conv3x3s)) 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): x[i] = self.branches[i](x[i]) x_fuse = [] for i in range(len(self.fuse_layers)): y = x[0] if i == 0 else self.fuse_layers[i][0](x[0]) for j in range(1, self.num_branches): if i == j: y = y + x[j] elif j > i: width_output = x[i].shape[-1] height_output = x[i].shape[-2] y = y + F.interpolate( self.fuse_layers[i][j](x[j]), size=[height_output, width_output], mode='bilinear', align_corners=True) else: y = y + self.fuse_layers[i][j](x[j]) x_fuse.append(self.relu(y)) return x_fuse blocks_dict = { 'BASIC': BasicBlock, 'BOTTLENECK': Bottleneck } class HighResolutionNet(nn.Module): def __init__(self, cfg, bn_type, bn_momentum, **kwargs): self.inplanes = 64 super(HighResolutionNet, self).__init__() if os.environ.get('full_res_stem'): Log.info("using full-resolution stem with stride=1") stem_stride = 1 self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=stem_stride, padding=1, bias=False) self.bn1 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(64, momentum=bn_momentum) self.relu = nn.ReLU(inplace=False) self.layer1 = self._make_layer(Bottleneck, 64, 64, 4, bn_type=bn_type, bn_momentum=bn_momentum) else: stem_stride = 2 self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=stem_stride, padding=1, bias=False) self.bn1 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(64, momentum=bn_momentum) self.conv2 = nn.Conv2d(64, 64, kernel_size=3, stride=stem_stride, padding=1, bias=False) self.bn2 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(64, momentum=bn_momentum) self.relu = nn.ReLU(inplace=False) self.layer1 = self._make_layer(Bottleneck, 64, 64, 4, bn_type=bn_type, bn_momentum=bn_momentum) self.stage2_cfg = cfg['STAGE2'] num_channels = self.stage2_cfg['NUM_CHANNELS'] block = blocks_dict[self.stage2_cfg['BLOCK']] num_channels = [ num_channels[i] * block.expansion for i in range(len(num_channels)) ] self.transition1 = self._make_transition_layer([256], num_channels, bn_type=bn_type, bn_momentum=bn_momentum) self.stage2, pre_stage_channels = self._make_stage( self.stage2_cfg, num_channels, bn_type=bn_type, bn_momentum=bn_momentum) self.stage3_cfg = cfg['STAGE3'] num_channels = self.stage3_cfg['NUM_CHANNELS'] block = blocks_dict[self.stage3_cfg['BLOCK']] num_channels = [ num_channels[i] * block.expansion for i in range(len(num_channels)) ] self.transition2 = self._make_transition_layer( pre_stage_channels, num_channels, bn_type=bn_type, bn_momentum=bn_momentum) self.stage3, pre_stage_channels = self._make_stage( self.stage3_cfg, num_channels, bn_type=bn_type, bn_momentum=bn_momentum) self.stage4_cfg = cfg['STAGE4'] num_channels = self.stage4_cfg['NUM_CHANNELS'] block = blocks_dict[self.stage4_cfg['BLOCK']] num_channels = [ num_channels[i] * block.expansion for i in range(len(num_channels)) ] self.transition3 = self._make_transition_layer( pre_stage_channels, num_channels, bn_type=bn_type, bn_momentum=bn_momentum) self.stage4, pre_stage_channels = self._make_stage( self.stage4_cfg, num_channels, multi_scale_output=True, bn_type=bn_type, bn_momentum=bn_momentum) if os.environ.get('keep_imagenet_head'): self.incre_modules, self.downsamp_modules, \ self.final_layer = self._make_head(pre_stage_channels, bn_type=bn_type, bn_momentum=bn_momentum) def _make_head(self, pre_stage_channels, bn_type, bn_momentum): head_block = Bottleneck head_channels = [32, 64, 128, 256] Log.info("pre_stage_channels: {}".format(pre_stage_channels)) Log.info("head_channels: {}".format(head_channels)) # Increasing the #channels on each resolution # from C, 2C, 4C, 8C to 128, 256, 512, 1024 incre_modules = [] for i, channels in enumerate(pre_stage_channels): incre_module = self._make_layer(head_block, channels, head_channels[i], 1, bn_type=bn_type, bn_momentum=bn_momentum ) incre_modules.append(incre_module) incre_modules = nn.ModuleList(incre_modules) # downsampling modules downsamp_modules = [] for i in range(len(pre_stage_channels) - 1): in_channels = head_channels[i] * head_block.expansion out_channels = head_channels[i + 1] * head_block.expansion downsamp_module = nn.Sequential( nn.Conv2d(in_channels=in_channels, out_channels=out_channels, kernel_size=3, stride=2, padding=1), ModuleHelper.BatchNorm2d(bn_type=bn_type)(out_channels, momentum=bn_momentum), nn.ReLU(inplace=False) ) downsamp_modules.append(downsamp_module) downsamp_modules = nn.ModuleList(downsamp_modules) final_layer = nn.Sequential( nn.Conv2d( in_channels=head_channels[3] * head_block.expansion, out_channels=2048, kernel_size=1, stride=1, padding=0 ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(2048, momentum=bn_momentum), nn.ReLU(inplace=False) ) return incre_modules, downsamp_modules, final_layer def _make_transition_layer( self, num_channels_pre_layer, num_channels_cur_layer, bn_type, bn_momentum): num_branches_cur = len(num_channels_cur_layer) num_branches_pre = len(num_channels_pre_layer) 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, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(num_channels_cur_layer[i], momentum=bn_momentum), nn.ReLU(inplace=False) ) ) else: transition_layers.append(None) else: conv3x3s = [] for j in range(i + 1 - num_branches_pre): inchannels = num_channels_pre_layer[-1] outchannels = num_channels_cur_layer[i] \ if j == i - num_branches_pre else inchannels conv3x3s.append( nn.Sequential( nn.Conv2d( inchannels, outchannels, 3, 2, 1, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(outchannels, momentum=bn_momentum), nn.ReLU(inplace=False) ) ) transition_layers.append(nn.Sequential(*conv3x3s)) return nn.ModuleList(transition_layers) def _make_layer(self, block, inplanes, planes, blocks, stride=1, bn_type=None, bn_momentum=0.1): downsample = None if stride != 1 or inplanes != planes * block.expansion: downsample = nn.Sequential( nn.Conv2d( inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(planes * block.expansion, momentum=bn_momentum) ) layers = [] layers.append(block(inplanes, planes, stride, downsample, bn_type=bn_type, bn_momentum=bn_momentum)) inplanes = planes * block.expansion for i in range(1, blocks): layers.append(block(inplanes, planes, bn_type=bn_type, bn_momentum=bn_momentum)) return nn.Sequential(*layers) def _make_stage(self, layer_config, num_inchannels, multi_scale_output=True, bn_type=None, bn_momentum=0.1): num_modules = layer_config['NUM_MODULES'] num_branches = layer_config['NUM_BRANCHES'] num_blocks = layer_config['NUM_BLOCKS'] num_channels = layer_config['NUM_CHANNELS'] block = blocks_dict[layer_config['BLOCK']] fuse_method = layer_config['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: reset_multi_scale_output = False else: reset_multi_scale_output = True modules.append( HighResolutionModule( num_branches, block, num_blocks, num_inchannels, num_channels, fuse_method, reset_multi_scale_output, bn_type, bn_momentum ) ) num_inchannels = modules[-1].get_num_inchannels() return nn.Sequential(*modules), num_inchannels def forward(self, x): if os.environ.get('full_res_stem'): x = self.conv1(x) x = self.bn1(x) x = self.relu(x) else: 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_list = [] for i in range(self.stage2_cfg['NUM_BRANCHES']): if self.transition1[i] is not None: x_list.append(self.transition1[i](x)) else: x_list.append(x) y_list = self.stage2(x_list) x_list = [] for i in range(self.stage3_cfg['NUM_BRANCHES']): if self.transition2[i] is not None: x_list.append(self.transition2[i](y_list[-1])) else: x_list.append(y_list[i]) y_list = self.stage3(x_list) if os.environ.get('drop_stage4'): return y_list x_list = [] for i in range(self.stage4_cfg['NUM_BRANCHES']): if self.transition3[i] is not None: x_list.append(self.transition3[i](y_list[-1])) else: x_list.append(y_list[i]) y_list = self.stage4(x_list) if os.environ.get('keep_imagenet_head'): # Classification Head x_list = [] y = self.incre_modules[0](y_list[0]) x_list.append(y) for i in range(len(self.downsamp_modules)): y = self.incre_modules[i + 1](y_list[i + 1]) + \ self.downsamp_modules[i](y) x_list.append(y) y = self.final_layer(y) del x_list[-1] x_list.append(y) return x_list return y_list class HighResolutionNext(nn.Module): def __init__(self, cfg, bn_type, **kwargs): super(HighResolutionNext, self).__init__() # stem net self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=2, padding=1, bias=False) self.bn1 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(64) self.relu = nn.ReLU(relu_inplace) self.stage1_cfg = cfg['STAGE1'] num_channels = self.stage1_cfg['NUM_CHANNELS'] block = blocks_dict[self.stage1_cfg['BLOCK']] num_channels = [ num_channels[i] * block.expansion for i in range(len(num_channels))] self.transition0 = self._make_transition_layer([64], num_channels, bn_type=bn_type) self.stage1, pre_stage_channels = self._make_stage( self.stage1_cfg, num_channels, bn_type=bn_type) self.stage2_cfg = cfg['STAGE2'] num_channels = self.stage2_cfg['NUM_CHANNELS'] block = blocks_dict[self.stage2_cfg['BLOCK']] num_channels = [ num_channels[i] * block.expansion for i in range(len(num_channels))] self.transition1 = self._make_transition_layer( pre_stage_channels, num_channels, bn_type=bn_type) self.stage2, pre_stage_channels = self._make_stage( self.stage2_cfg, num_channels, bn_type=bn_type) self.stage3_cfg = cfg['STAGE3'] num_channels = self.stage3_cfg['NUM_CHANNELS'] block = blocks_dict[self.stage3_cfg['BLOCK']] num_channels = [ num_channels[i] * block.expansion for i in range(len(num_channels))] self.transition2 = self._make_transition_layer( pre_stage_channels, num_channels, bn_type=bn_type) self.stage3, pre_stage_channels = self._make_stage( self.stage3_cfg, num_channels, bn_type=bn_type) self.stage4_cfg = cfg['STAGE4'] num_channels = self.stage4_cfg['NUM_CHANNELS'] block = blocks_dict[self.stage4_cfg['BLOCK']] num_channels = [ num_channels[i] * block.expansion for i in range(len(num_channels))] self.transition3 = self._make_transition_layer( pre_stage_channels, num_channels, bn_type=bn_type) self.stage4, pre_stage_channels = self._make_stage( self.stage4_cfg, num_channels, multi_scale_output=True, bn_type=bn_type) def _make_transition_layer( self, num_channels_pre_layer, num_channels_cur_layer, bn_type): num_branches_cur = len(num_channels_cur_layer) num_branches_pre = len(num_channels_pre_layer) 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, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(num_channels_cur_layer[i]), nn.ReLU(relu_inplace) ) ) else: transition_layers.append(None) else: conv3x3s = [] for j in range(i + 1 - num_branches_pre): inchannels = num_channels_pre_layer[-1] outchannels = num_channels_cur_layer[i] \ if j == i - num_branches_pre else inchannels conv3x3s.append( nn.Sequential( nn.Conv2d( inchannels, outchannels, 3, 2, 1, bias=False ), ModuleHelper.BatchNorm2d(bn_type=bn_type)(outchannels), nn.ReLU(relu_inplace) ) ) transition_layers.append(nn.Sequential(*conv3x3s)) return nn.ModuleList(transition_layers) def _make_stage(self, layer_config, num_inchannels, multi_scale_output=True, bn_type=None): num_modules = layer_config['NUM_MODULES'] num_branches = layer_config['NUM_BRANCHES'] num_blocks = layer_config['NUM_BLOCKS'] num_channels = layer_config['NUM_CHANNELS'] block = blocks_dict[layer_config['BLOCK']] fuse_method = layer_config['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: reset_multi_scale_output = False else: reset_multi_scale_output = True modules.append( HighResolutionModule( num_branches, block, num_blocks, num_inchannels, num_channels, fuse_method, reset_multi_scale_output, bn_type ) ) num_inchannels = modules[-1].get_num_inchannels() return nn.Sequential(*modules), num_inchannels def forward(self, x): x = self.conv1(x) x = self.bn1(x) x = self.relu(x) x_list = [] for i in range(self.stage1_cfg['NUM_BRANCHES']): if self.transition0[i] is not None: x_list.append(self.transition0[i](x)) else: x_list.append(x) y_list = self.stage1(x_list) x_list = [] for i in range(self.stage2_cfg['NUM_BRANCHES']): if self.transition1[i] is not None: if i == 0: x_list.append(self.transition1[i](y_list[0])) else: x_list.append(self.transition1[i](y_list[-1])) else: x_list.append(y_list[i]) y_list = self.stage2(x_list) x_list = [] for i in range(self.stage3_cfg['NUM_BRANCHES']): if self.transition2[i] is not None: x_list.append(self.transition2[i](y_list[-1])) else: x_list.append(y_list[i]) y_list = self.stage3(x_list) x_list = [] for i in range(self.stage4_cfg['NUM_BRANCHES']): if self.transition3[i] is not None: x_list.append(self.transition3[i](y_list[-1])) else: x_list.append(y_list[i]) x = self.stage4(x_list) return x class HRNetBackbone(object): def __init__(self, configer): self.configer = configer def __call__(self): arch = self.configer.get('network', 'backbone') resume = self.configer.get('network', 'resume') from lib.models.backbones.hrnet.hrnet_config import MODEL_CONFIGS if arch == 'hrnet18': arch_net = HighResolutionNet(MODEL_CONFIGS['hrnet18'], bn_type='torchsyncbn', bn_momentum=0.1) if resume is None: arch_net = ModuleHelper.load_model(arch_net, pretrained=self.configer.get('network', 'pretrained'), all_match=False, network='hrnet') elif arch == 'hrnet32': arch_net = HighResolutionNet(MODEL_CONFIGS['hrnet32'], bn_type='torchsyncbn', bn_momentum=0.1) if resume is None: arch_net = ModuleHelper.load_model(arch_net, pretrained=self.configer.get('network', 'pretrained'), all_match=False, network='hrnet') elif arch == 'hrnet48': arch_net = HighResolutionNet(MODEL_CONFIGS['hrnet48'], bn_type='torchsyncbn', bn_momentum=0.1) if resume is None: arch_net = ModuleHelper.load_model(arch_net, pretrained=self.configer.get('network', 'pretrained'), all_match=False, network='hrnet') elif arch == 'hrnet64': arch_net = HighResolutionNet(MODEL_CONFIGS['hrnet64'], bn_type='torchsyncbn', bn_momentum=0.1) if resume is None: arch_net = ModuleHelper.load_model(arch_net, pretrained=self.configer.get('network', 'pretrained'), all_match=False, network='hrnet') elif arch == 'hrnet2x20': arch_net = HighResolutionNext(MODEL_CONFIGS['hrnet2x20'], bn_type=self.configer.get('network', 'bn_type')) if resume is None: arch_net = ModuleHelper.load_model(arch_net, pretrained=self.configer.get('network', 'pretrained'), all_match=False, network='hrnet') else: raise Exception('Architecture undefined!') return arch_net