data/lib/models/nets/hrnet.py

##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
## Created by: RainbowSecret
## Microsoft Research
## yuyua@microsoft.com
## Copyright (c) 2018
##
## This source code is licensed under the MIT-style license found in the
## LICENSE file in the root directory of this source tree 
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

import os
import pdb
import torch
import torch.nn as nn
import torch.nn.functional as F

from lib.models.backbones.backbone_selector import BackboneSelector
from lib.models.tools.module_helper import ModuleHelper
from lib.models.modules.projection import ProjectionHead
from lib.utils.tools.logger import Logger as Log
from lib.models.modules.hanet_attention import HANet_Conv


class HRNet_W48(nn.Module):
    """
    deep high-resolution representation learning for human pose estimation, CVPR2019
    """

    def __init__(self, configer):
        super(HRNet_W48, self).__init__()
        self.configer = configer
        self.num_classes = self.configer.get('data', 'num_classes')
        self.backbone = BackboneSelector(configer).get_backbone()

        # extra added layers
        in_channels = 720  # 48 + 96 + 192 + 384
        self.cls_head = nn.Sequential(
            nn.Conv2d(in_channels, in_channels, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(in_channels, bn_type=self.configer.get('network', 'bn_type')),
            nn.Dropout2d(0.10),
            nn.Conv2d(in_channels, self.num_classes, kernel_size=1, stride=1, padding=0, bias=False)
        )

    def forward(self, x_):
        x = self.backbone(x_)
        _, _, h, w = x[0].size()

        feat1 = x[0]
        feat2 = F.interpolate(x[1], size=(h, w), mode="bilinear", align_corners=True)
        feat3 = F.interpolate(x[2], size=(h, w), mode="bilinear", align_corners=True)
        feat4 = F.interpolate(x[3], size=(h, w), mode="bilinear", align_corners=True)

        feats = torch.cat([feat1, feat2, feat3, feat4], 1)
        out = self.cls_head(feats)
        out = F.interpolate(out, size=(x_.size(2), x_.size(3)), mode="bilinear", align_corners=True)
        return out


class HRNet_W48_CONTRAST(nn.Module):
    """
    deep high-resolution representation learning for human pose estimation, CVPR2019
    """

    def __init__(self, configer):
        super(HRNet_W48_CONTRAST, self).__init__()
        self.configer = configer
        self.num_classes = self.configer.get('data', 'num_classes')
        self.backbone = BackboneSelector(configer).get_backbone()
        self.proj_dim = self.configer.get('contrast', 'proj_dim')

        # extra added layers
        in_channels = 720  # 48 + 96 + 192 + 384
        self.cls_head = nn.Sequential(
            nn.Conv2d(in_channels, in_channels, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(in_channels, bn_type=self.configer.get('network', 'bn_type')),
            nn.Dropout2d(0.10),
            nn.Conv2d(in_channels, self.num_classes, kernel_size=1, stride=1, padding=0, bias=False)
        )

        self.proj_head = ProjectionHead(dim_in=in_channels, proj_dim=self.proj_dim)

    def forward(self, x_, with_embed=False, is_eval=False):
        x = self.backbone(x_)
        _, _, h, w = x[0].size()

        feat1 = x[0]
        feat2 = F.interpolate(x[1], size=(h, w), mode="bilinear", align_corners=True)
        feat3 = F.interpolate(x[2], size=(h, w), mode="bilinear", align_corners=True)
        feat4 = F.interpolate(x[3], size=(h, w), mode="bilinear", align_corners=True)

        feats = torch.cat([feat1, feat2, feat3, feat4], 1)
        out = self.cls_head(feats)

        emb = self.proj_head(feats)
        return {'seg': out, 'embed': emb}


class HRNet_W48_OCR_CONTRAST(nn.Module):
    def __init__(self, configer):
        super(HRNet_W48_OCR_CONTRAST, self).__init__()
        self.configer = configer
        self.num_classes = self.configer.get('data', 'num_classes')
        self.backbone = BackboneSelector(configer).get_backbone()
        self.proj_dim = self.configer.get('contrast', 'proj_dim')

        in_channels = 720
        self.conv3x3 = nn.Sequential(
            nn.Conv2d(in_channels, 512, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(512, bn_type=self.configer.get('network', 'bn_type')),
        )
        from lib.models.modules.spatial_ocr_block import SpatialGather_Module
        self.ocr_gather_head = SpatialGather_Module(self.num_classes)
        from lib.models.modules.spatial_ocr_block import SpatialOCR_Module
        self.ocr_distri_head = SpatialOCR_Module(in_channels=512,
                                                 key_channels=256,
                                                 out_channels=512,
                                                 scale=1,
                                                 dropout=0.05,
                                                 bn_type=self.configer.get('network', 'bn_type'))
        self.cls_head = nn.Conv2d(512, self.num_classes, kernel_size=1, stride=1, padding=0, bias=True)
        self.aux_head = nn.Sequential(
            nn.Conv2d(in_channels, in_channels, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(in_channels, bn_type=self.configer.get('network', 'bn_type')),
            nn.Conv2d(in_channels, self.num_classes, kernel_size=1, stride=1, padding=0, bias=True)
        )

        self.proj_head = ProjectionHead(dim_in=in_channels, proj_dim=self.proj_dim)

    def forward(self, x_, with_embed=False, is_eval=False):
        x = self.backbone(x_)
        _, _, h, w = x[0].size()

        feat1 = x[0]
        feat2 = F.interpolate(x[1], size=(h, w), mode="bilinear", align_corners=True)
        feat3 = F.interpolate(x[2], size=(h, w), mode="bilinear", align_corners=True)
        feat4 = F.interpolate(x[3], size=(h, w), mode="bilinear", align_corners=True)

        feats = torch.cat([feat1, feat2, feat3, feat4], 1)
        out_aux = self.aux_head(feats)

        emb = self.proj_head(feats)

        feats = self.conv3x3(feats)

        context = self.ocr_gather_head(feats, out_aux)
        feats = self.ocr_distri_head(feats, context)

        out = self.cls_head(feats)

        return {'seg': out, 'seg_aux': out_aux, 'embed': emb}


class HRNet_W48_MEM(nn.Module):
    def __init__(self, configer, dim=256, m=0.999, with_masked_ppm=False):
        super(HRNet_W48_MEM, self).__init__()
        self.configer = configer
        self.m = m
        self.r = self.configer.get('contrast', 'memory_size')
        self.with_masked_ppm = with_masked_ppm

        num_classes = self.configer.get('data', 'num_classes')

        self.encoder_q = HRNet_W48_CONTRAST(configer)

        self.register_buffer("segment_queue", torch.randn(num_classes, self.r, dim))
        self.segment_queue = nn.functional.normalize(self.segment_queue, p=2, dim=2)
        self.register_buffer("segment_queue_ptr", torch.zeros(num_classes, dtype=torch.long))

        self.register_buffer("pixel_queue", torch.randn(num_classes, self.r, dim))
        self.pixel_queue = nn.functional.normalize(self.pixel_queue, p=2, dim=2)
        self.register_buffer("pixel_queue_ptr", torch.zeros(num_classes, dtype=torch.long))

    @torch.no_grad()
    def _momentum_update_key_encoder(self):
        for param_q, param_k in zip(self.encoder_q.parameters(), self.encoder_k.parameters()):
            param_k.data = param_k.data * self.m + param_q.data * (1. - self.m)

    def forward(self, im_q, lb_q=None, with_embed=True, is_eval=False):
        if is_eval is True or lb_q is None:
            ret = self.encoder_q(im_q, with_embed=with_embed)
            return ret

        ret = self.encoder_q(im_q)

        q = ret['embed']
        out = ret['seg']

        return {'seg': out, 'embed': q, 'key': q.detach(), 'lb_key': lb_q.detach()}


class HRNet_W48_OCR(nn.Module):
    def __init__(self, configer):
        super(HRNet_W48_OCR, self).__init__()
        self.configer = configer
        self.num_classes = self.configer.get('data', 'num_classes')
        self.backbone = BackboneSelector(configer).get_backbone()

        in_channels = 720
        self.conv3x3 = nn.Sequential(
            nn.Conv2d(in_channels, 512, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(512, bn_type=self.configer.get('network', 'bn_type')),
        )
        from lib.models.modules.spatial_ocr_block import SpatialGather_Module
        self.ocr_gather_head = SpatialGather_Module(self.num_classes)
        from lib.models.modules.spatial_ocr_block import SpatialOCR_Module
        self.ocr_distri_head = SpatialOCR_Module(in_channels=512,
                                                 key_channels=256,
                                                 out_channels=512,
                                                 scale=1,
                                                 dropout=0.05,
                                                 bn_type=self.configer.get('network', 'bn_type'))
        self.cls_head = nn.Conv2d(512, self.num_classes, kernel_size=1, stride=1, padding=0, bias=True)
        self.aux_head = nn.Sequential(
            nn.Conv2d(in_channels, in_channels, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(in_channels, bn_type=self.configer.get('network', 'bn_type')),
            nn.Conv2d(in_channels, self.num_classes, kernel_size=1, stride=1, padding=0, bias=True)
        )

    def forward(self, x_):
        x = self.backbone(x_)
        _, _, h, w = x[0].size()

        feat1 = x[0]
        feat2 = F.interpolate(x[1], size=(h, w), mode="bilinear", align_corners=True)
        feat3 = F.interpolate(x[2], size=(h, w), mode="bilinear", align_corners=True)
        feat4 = F.interpolate(x[3], size=(h, w), mode="bilinear", align_corners=True)

        feats = torch.cat([feat1, feat2, feat3, feat4], 1)
        out_aux = self.aux_head(feats)

        feats = self.conv3x3(feats)

        context = self.ocr_gather_head(feats, out_aux)
        feats = self.ocr_distri_head(feats, context)

        out = self.cls_head(feats)

        out_aux = F.interpolate(out_aux, size=(x_.size(2), x_.size(3)), mode="bilinear", align_corners=True)
        out = F.interpolate(out, size=(x_.size(2), x_.size(3)), mode="bilinear", align_corners=True)
        return out_aux, out


class HRNet_W48_OCR_B(nn.Module):
    """
    Considering that the 3x3 convolution on the 4x resolution feature map is expensive,
    we can decrease the intermediate channels from 512 to 256 w/o performance loss.
    """

    def __init__(self, configer):
        super(HRNet_W48_OCR_B, self).__init__()
        self.configer = configer
        self.num_classes = self.configer.get('data', 'num_classes')
        self.backbone = BackboneSelector(configer).get_backbone()

        in_channels = 720  # 48 + 96 + 192 + 384
        self.conv3x3 = nn.Sequential(
            nn.Conv2d(in_channels, 256, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(256, bn_type=self.configer.get('network', 'bn_type')),
        )
        from lib.models.modules.spatial_ocr_block import SpatialGather_Module
        self.ocr_gather_head = SpatialGather_Module(self.num_classes)
        from lib.models.modules.spatial_ocr_block import SpatialOCR_Module
        self.ocr_distri_head = SpatialOCR_Module(in_channels=256,
                                                 key_channels=128,
                                                 out_channels=256,
                                                 scale=1,
                                                 dropout=0.05,
                                                 bn_type=self.configer.get('network', 'bn_type'))

        self.cls_head = nn.Conv2d(256, self.num_classes, kernel_size=1, stride=1, padding=0, bias=True)
        self.aux_head = nn.Sequential(
            nn.Conv2d(in_channels, 256, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(256, bn_type=self.configer.get('network', 'bn_type')),
            nn.Conv2d(256, self.num_classes, kernel_size=1, stride=1, padding=0, bias=True)
        )

    def forward(self, x_):
        x = self.backbone(x_)
        _, _, h, w = x[0].size()

        feat1 = x[0]
        feat2 = F.interpolate(x[1], size=(h, w), mode="bilinear", align_corners=True)
        feat3 = F.interpolate(x[2], size=(h, w), mode="bilinear", align_corners=True)
        feat4 = F.interpolate(x[3], size=(h, w), mode="bilinear", align_corners=True)

        feats = torch.cat([feat1, feat2, feat3, feat4], 1)
        out_aux = self.aux_head(feats)

        feats = self.conv3x3(feats)

        context = self.ocr_gather_head(feats, out_aux)
        feats = self.ocr_distri_head(feats, context)

        out = self.cls_head(feats)

        out_aux = F.interpolate(out_aux, size=(x_.size(2), x_.size(3)), mode="bilinear", align_corners=True)
        out = F.interpolate(out, size=(x_.size(2), x_.size(3)), mode="bilinear", align_corners=True)
        return out_aux, out


class HRNet_W48_OCR_B_HA(nn.Module):
    """
    Considering that the 3x3 convolution on the 4x resolution feature map is expensive,
    we can decrease the intermediate channels from 512 to 256 w/o performance loss.
    """

    def __init__(self, configer):
        super(HRNet_W48_OCR_B_HA, self).__init__()
        self.configer = configer
        self.num_classes = self.configer.get('data', 'num_classes')
        self.backbone = BackboneSelector(configer).get_backbone()

        in_channels = 720  # 48 + 96 + 192 + 384
        self.conv3x3 = nn.Sequential(
            nn.Conv2d(in_channels, 256, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(256, bn_type=self.configer.get('network', 'bn_type')),
        )
        from lib.models.modules.spatial_ocr_block import SpatialGather_Module
        self.ocr_gather_head = SpatialGather_Module(self.num_classes)
        from lib.models.modules.spatial_ocr_block import SpatialOCR_Module
        self.ocr_distri_head = SpatialOCR_Module(in_channels=256,
                                                 key_channels=128,
                                                 out_channels=256,
                                                 scale=1,
                                                 dropout=0.05,
                                                 bn_type=self.configer.get('network', 'bn_type'))
        self.cls_head = nn.Conv2d(256, self.num_classes, kernel_size=1, stride=1, padding=0, bias=True)
        self.aux_head = nn.Sequential(
            nn.Conv2d(in_channels, 256, kernel_size=3, stride=1, padding=1),
            ModuleHelper.BNReLU(256, bn_type=self.configer.get('network', 'bn_type')),
            nn.Conv2d(256, self.num_classes, kernel_size=1, stride=1, padding=0, bias=True)
        )

        self.ha1 = HANet_Conv(384, 384, bn_type=self.configer.get('network', 'bn_type'))
        self.ha2 = HANet_Conv(192, 192, bn_type=self.configer.get('network', 'bn_type'))
        self.ha3 = HANet_Conv(96, 96, bn_type=self.configer.get('network', 'bn_type'))
        self.ha4 = HANet_Conv(48, 48, bn_type=self.configer.get('network', 'bn_type'))

    def forward(self, x_):
        x = self.backbone(x_)
        _, _, h, w = x[0].size()

        x[0] = x[0] + self.ha1(x[0])
        x[1] = x[1] + self.ha1(x[1])
        x[2] = x[2] + self.ha1(x[2])
        x[3] = x[3] + self.ha1(x[3])

        feat1 = x[0]
        feat2 = F.interpolate(x[1], size=(h, w), mode="bilinear", align_corners=True)
        feat3 = F.interpolate(x[2], size=(h, w), mode="bilinear", align_corners=True)
        feat4 = F.interpolate(x[3], size=(h, w), mode="bilinear", align_corners=True)

        feats = torch.cat([feat1, feat2, feat3, feat4], 1)
        out_aux = self.aux_head(feats)

        feats = self.conv3x3(feats)

        context = self.ocr_gather_head(feats, out_aux)
        feats = self.ocr_distri_head(feats, context)

        out = self.cls_head(feats)

        out_aux = F.interpolate(out_aux, size=(x_.size(2), x_.size(3)), mode="bilinear", align_corners=True)
        out = F.interpolate(out, size=(x_.size(2), x_.size(3)), mode="bilinear", align_corners=True)
        return out_aux, out