third_party/lanet/network_v1/modules.py

from curses import is_term_resized
import torch
import torch.nn as nn
import torch.nn.functional as F

from torchvision import models
from lanet_utils import image_grid


class ConvBlock(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(ConvBlock, self).__init__()

        self.conv = nn.Sequential(
            nn.Conv2d(
                in_channels,
                out_channels,
                kernel_size=3,
                stride=1,
                padding=1,
                bias=False,
            ),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True),
            nn.Conv2d(
                out_channels,
                out_channels,
                kernel_size=3,
                stride=1,
                padding=1,
                bias=False,
            ),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True),
        )

    def forward(self, x):
        return self.conv(x)


class DilationConv3x3(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(DilationConv3x3, self).__init__()

        self.conv = nn.Conv2d(
            in_channels,
            out_channels,
            kernel_size=3,
            stride=1,
            padding=2,
            dilation=2,
            bias=False,
        )
        self.bn = nn.BatchNorm2d(out_channels)

    def forward(self, x):
        x = self.conv(x)
        x = self.bn(x)
        return x


class InterestPointModule(nn.Module):
    def __init__(self, is_test=False):
        super(InterestPointModule, self).__init__()
        self.is_test = is_test

        model = models.vgg16_bn(pretrained=True)

        # use the first 23 layers as encoder
        self.encoder = nn.Sequential(*list(model.features.children())[:33])

        # score head
        self.score_head = nn.Sequential(
            nn.Conv2d(512, 256, kernel_size=3, stride=1, padding=1, bias=False),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256, 4, kernel_size=3, stride=1, padding=1),
        )
        self.softmax = nn.Softmax(dim=1)

        # location head
        self.loc_head = nn.Sequential(
            nn.Conv2d(512, 256, kernel_size=3, stride=1, padding=1, bias=False),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
        )
        # location out
        self.loc_out = nn.Conv2d(256, 2, kernel_size=3, stride=1, padding=1)
        self.shift_out = nn.Conv2d(256, 1, kernel_size=3, stride=1, padding=1)

        # descriptor out
        self.des_out2 = DilationConv3x3(128, 256)
        self.des_out3 = DilationConv3x3(256, 256)
        self.des_out4 = DilationConv3x3(512, 256)

    def forward(self, x):
        B, _, H, W = x.shape

        x = self.encoder[2](self.encoder[1](self.encoder[0](x)))
        x = self.encoder[5](self.encoder[4](self.encoder[3](x)))

        x = self.encoder[6](x)
        x = self.encoder[9](self.encoder[8](self.encoder[7](x)))
        x2 = self.encoder[12](self.encoder[11](self.encoder[10](x)))

        x = self.encoder[13](x2)
        x = self.encoder[16](self.encoder[15](self.encoder[14](x)))
        x = self.encoder[19](self.encoder[18](self.encoder[17](x)))
        x3 = self.encoder[22](self.encoder[21](self.encoder[20](x)))

        x = self.encoder[23](x3)
        x = self.encoder[26](self.encoder[25](self.encoder[24](x)))
        x = self.encoder[29](self.encoder[28](self.encoder[27](x)))
        x = self.encoder[32](self.encoder[31](self.encoder[30](x)))

        B, _, Hc, Wc = x.shape

        # score head
        score_x = self.score_head(x)
        aware = self.softmax(score_x[:, 0:3, :, :])
        score = score_x[:, 3, :, :].unsqueeze(1).sigmoid()

        border_mask = torch.ones(B, Hc, Wc)
        border_mask[:, 0] = 0
        border_mask[:, Hc - 1] = 0
        border_mask[:, :, 0] = 0
        border_mask[:, :, Wc - 1] = 0
        border_mask = border_mask.unsqueeze(1)
        score = score * border_mask.to(score.device)

        # location head
        coord_x = self.loc_head(x)
        coord_cell = self.loc_out(coord_x).tanh()

        shift_ratio = self.shift_out(coord_x).sigmoid() * 2.0

        step = ((H / Hc) - 1) / 2.0
        center_base = (
            image_grid(
                B,
                Hc,
                Wc,
                dtype=coord_cell.dtype,
                device=coord_cell.device,
                ones=False,
                normalized=False,
            ).mul(H / Hc)
            + step
        )

        coord_un = center_base.add(coord_cell.mul(shift_ratio * step))
        coord = coord_un.clone()
        coord[:, 0] = torch.clamp(coord_un[:, 0], min=0, max=W - 1)
        coord[:, 1] = torch.clamp(coord_un[:, 1], min=0, max=H - 1)

        # descriptor block
        desc_block = []
        desc_block.append(self.des_out2(x2))
        desc_block.append(self.des_out3(x3))
        desc_block.append(self.des_out4(x))
        desc_block.append(aware)

        if self.is_test:
            coord_norm = coord[:, :2].clone()
            coord_norm[:, 0] = (coord_norm[:, 0] / (float(W - 1) / 2.0)) - 1.0
            coord_norm[:, 1] = (coord_norm[:, 1] / (float(H - 1) / 2.0)) - 1.0
            coord_norm = coord_norm.permute(0, 2, 3, 1)

            desc2 = torch.nn.functional.grid_sample(desc_block[0], coord_norm)
            desc3 = torch.nn.functional.grid_sample(desc_block[1], coord_norm)
            desc4 = torch.nn.functional.grid_sample(desc_block[2], coord_norm)
            aware = desc_block[3]

            desc = (
                torch.mul(desc2, aware[:, 0, :, :])
                + torch.mul(desc3, aware[:, 1, :, :])
                + torch.mul(desc4, aware[:, 2, :, :])
            )
            desc = desc.div(
                torch.unsqueeze(torch.norm(desc, p=2, dim=1), 1)
            )  # Divide by norm to normalize.

            return score, coord, desc

        return score, coord, desc_block


class CorrespondenceModule(nn.Module):
    def __init__(self, match_type="dual_softmax"):
        super(CorrespondenceModule, self).__init__()
        self.match_type = match_type

        if self.match_type == "dual_softmax":
            self.temperature = 0.1
        else:
            raise NotImplementedError()

    def forward(self, source_desc, target_desc):
        b, c, h, w = source_desc.size()

        source_desc = source_desc.div(
            torch.unsqueeze(torch.norm(source_desc, p=2, dim=1), 1)
        ).view(b, -1, h * w)
        target_desc = target_desc.div(
            torch.unsqueeze(torch.norm(target_desc, p=2, dim=1), 1)
        ).view(b, -1, h * w)

        if self.match_type == "dual_softmax":
            sim_mat = (
                torch.einsum("bcm, bcn -> bmn", source_desc, target_desc)
                / self.temperature
            )
            confidence_matrix = F.softmax(sim_mat, 1) * F.softmax(sim_mat, 2)
        else:
            raise NotImplementedError()

        return confidence_matrix