2021-02-09 12:24:51 +01:00
|
|
|
import torch
|
|
|
|
import numpy as np
|
|
|
|
import torch.nn as nn
|
|
|
|
import torch.nn.functional as F
|
|
|
|
from model.warplayer import warp
|
|
|
|
|
|
|
|
|
|
|
|
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
|
|
|
|
|
|
def conv_wo_act(in_planes, out_planes, kernel_size=3, stride=1, padding=1, dilation=1):
|
|
|
|
return nn.Sequential(
|
|
|
|
nn.Conv2d(in_planes, out_planes, kernel_size=kernel_size, stride=stride,
|
|
|
|
padding=padding, dilation=dilation, bias=True),
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
def conv(in_planes, out_planes, kernel_size=3, stride=1, padding=1, dilation=1):
|
|
|
|
return nn.Sequential(
|
|
|
|
nn.Conv2d(in_planes, out_planes, kernel_size=kernel_size, stride=stride,
|
|
|
|
padding=padding, dilation=dilation, bias=True),
|
|
|
|
nn.PReLU(out_planes)
|
|
|
|
)
|
|
|
|
|
|
|
|
class IFBlock(nn.Module):
|
|
|
|
def __init__(self, in_planes, scale=1, c=64):
|
|
|
|
super(IFBlock, self).__init__()
|
|
|
|
self.scale = scale
|
|
|
|
self.conv0 = nn.Sequential(
|
|
|
|
conv(in_planes, c, 3, 2, 1),
|
|
|
|
conv(c, 2*c, 3, 2, 1),
|
|
|
|
)
|
|
|
|
self.convblock = nn.Sequential(
|
|
|
|
conv(2*c, 2*c),
|
|
|
|
conv(2*c, 2*c),
|
|
|
|
conv(2*c, 2*c),
|
|
|
|
conv(2*c, 2*c),
|
|
|
|
conv(2*c, 2*c),
|
|
|
|
conv(2*c, 2*c),
|
|
|
|
)
|
|
|
|
self.conv1 = nn.ConvTranspose2d(2*c, 4, 4, 2, 1)
|
|
|
|
|
2021-03-02 22:54:30 +01:00
|
|
|
def forward(self, x):
|
|
|
|
if self.scale != 1:
|
|
|
|
x = F.interpolate(x, scale_factor=1. / self.scale, mode="bilinear",
|
2021-02-09 12:24:51 +01:00
|
|
|
align_corners=False)
|
|
|
|
x = self.conv0(x)
|
|
|
|
x = self.convblock(x)
|
|
|
|
x = self.conv1(x)
|
|
|
|
flow = x
|
2021-03-02 22:54:30 +01:00
|
|
|
if self.scale != 1:
|
|
|
|
flow = F.interpolate(flow, scale_factor=self.scale, mode="bilinear",
|
|
|
|
align_corners=False)
|
|
|
|
return flow
|
2021-02-09 12:24:51 +01:00
|
|
|
|
|
|
|
|
|
|
|
class IFNet(nn.Module):
|
|
|
|
def __init__(self):
|
|
|
|
super(IFNet, self).__init__()
|
|
|
|
self.block0 = IFBlock(6, scale=8, c=192)
|
|
|
|
self.block1 = IFBlock(10, scale=4, c=128)
|
|
|
|
self.block2 = IFBlock(10, scale=2, c=96)
|
|
|
|
self.block3 = IFBlock(10, scale=1, c=48)
|
|
|
|
|
2021-06-15 15:41:15 +02:00
|
|
|
def forward(self, x, scale=1.0):
|
|
|
|
if scale != 1.0:
|
|
|
|
x = F.interpolate(x, scale_factor=scale, mode="bilinear", align_corners=False)
|
2021-03-02 22:54:30 +01:00
|
|
|
flow0 = self.block0(x)
|
2021-02-09 12:24:51 +01:00
|
|
|
F1 = flow0
|
2021-06-15 15:41:15 +02:00
|
|
|
F1_large = F.interpolate(F1, scale_factor=2.0, mode="bilinear", align_corners=False) * 2.0
|
2021-02-09 12:24:51 +01:00
|
|
|
warped_img0 = warp(x[:, :3], F1_large[:, :2])
|
|
|
|
warped_img1 = warp(x[:, 3:], F1_large[:, 2:4])
|
2021-03-02 22:54:30 +01:00
|
|
|
flow1 = self.block1(torch.cat((warped_img0, warped_img1, F1_large), 1))
|
2021-02-09 12:24:51 +01:00
|
|
|
F2 = (flow0 + flow1)
|
2021-06-15 15:41:15 +02:00
|
|
|
F2_large = F.interpolate(F2, scale_factor=2.0, mode="bilinear", align_corners=False) * 2.0
|
2021-02-09 12:24:51 +01:00
|
|
|
warped_img0 = warp(x[:, :3], F2_large[:, :2])
|
|
|
|
warped_img1 = warp(x[:, 3:], F2_large[:, 2:4])
|
2021-03-02 22:54:30 +01:00
|
|
|
flow2 = self.block2(torch.cat((warped_img0, warped_img1, F2_large), 1))
|
2021-02-09 12:24:51 +01:00
|
|
|
F3 = (flow0 + flow1 + flow2)
|
2021-06-15 15:41:15 +02:00
|
|
|
F3_large = F.interpolate(F3, scale_factor=2.0, mode="bilinear", align_corners=False) * 2.0
|
2021-02-09 12:24:51 +01:00
|
|
|
warped_img0 = warp(x[:, :3], F3_large[:, :2])
|
|
|
|
warped_img1 = warp(x[:, 3:], F3_large[:, 2:4])
|
2021-03-02 22:54:30 +01:00
|
|
|
flow3 = self.block3(torch.cat((warped_img0, warped_img1, F3_large), 1))
|
2021-02-09 12:24:51 +01:00
|
|
|
F4 = (flow0 + flow1 + flow2 + flow3)
|
2021-06-15 15:41:15 +02:00
|
|
|
if scale != 1.0:
|
|
|
|
F4 = F.interpolate(F4, scale_factor=1 / scale, mode="bilinear", align_corners=False) / scale
|
2021-02-09 12:24:51 +01:00
|
|
|
return F4, [F1, F2, F3, F4]
|
|
|
|
|
|
|
|
if __name__ == '__main__':
|
|
|
|
img0 = torch.zeros(3, 3, 256, 256).float().to(device)
|
|
|
|
img1 = torch.tensor(np.random.normal(
|
|
|
|
0, 1, (3, 3, 256, 256))).float().to(device)
|
|
|
|
imgs = torch.cat((img0, img1), 1)
|
|
|
|
flownet = IFNet()
|
|
|
|
flow, _ = flownet(imgs)
|
|
|
|
print(flow.shape)
|