File size: 3,419 Bytes
854f0d0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 |
import torch
# ! amazing!!!! autograd.grad with set_detect_anomaly(True) will cause memory leak
# ! https://github.com/pytorch/pytorch/issues/51349
# torch.autograd.set_detect_anomaly(True)
import torch.nn as nn
import torch.nn.functional as F
from inplace_abn import InPlaceABN
############################################# MVS Net models ################################################
class ConvBnReLU(nn.Module):
def __init__(self, in_channels, out_channels,
kernel_size=3, stride=1, pad=1,
norm_act=InPlaceABN):
super(ConvBnReLU, self).__init__()
self.conv = nn.Conv2d(in_channels, out_channels,
kernel_size, stride=stride, padding=pad, bias=False)
self.bn = norm_act(out_channels)
def forward(self, x):
return self.bn(self.conv(x))
class ConvBnReLU3D(nn.Module):
def __init__(self, in_channels, out_channels,
kernel_size=3, stride=1, pad=1,
norm_act=InPlaceABN):
super(ConvBnReLU3D, self).__init__()
self.conv = nn.Conv3d(in_channels, out_channels,
kernel_size, stride=stride, padding=pad, bias=False)
self.bn = norm_act(out_channels)
# self.bn = nn.ReLU()
def forward(self, x):
return self.bn(self.conv(x))
################################### feature net ######################################
class FeatureNet(nn.Module):
"""
output 3 levels of features using a FPN structure
"""
def __init__(self, norm_act=InPlaceABN):
super(FeatureNet, self).__init__()
self.conv0 = nn.Sequential(
ConvBnReLU(3, 8, 3, 1, 1, norm_act=norm_act),
ConvBnReLU(8, 8, 3, 1, 1, norm_act=norm_act))
self.conv1 = nn.Sequential(
ConvBnReLU(8, 16, 5, 2, 2, norm_act=norm_act),
ConvBnReLU(16, 16, 3, 1, 1, norm_act=norm_act),
ConvBnReLU(16, 16, 3, 1, 1, norm_act=norm_act))
self.conv2 = nn.Sequential(
ConvBnReLU(16, 32, 5, 2, 2, norm_act=norm_act),
ConvBnReLU(32, 32, 3, 1, 1, norm_act=norm_act),
ConvBnReLU(32, 32, 3, 1, 1, norm_act=norm_act))
self.toplayer = nn.Conv2d(32, 32, 1)
self.lat1 = nn.Conv2d(16, 32, 1)
self.lat0 = nn.Conv2d(8, 32, 1)
# to reduce channel size of the outputs from FPN
self.smooth1 = nn.Conv2d(32, 16, 3, padding=1)
self.smooth0 = nn.Conv2d(32, 8, 3, padding=1)
def _upsample_add(self, x, y):
return F.interpolate(x, scale_factor=2,
mode="bilinear", align_corners=True) + y
def forward(self, x):
# x: (B, 3, H, W)
conv0 = self.conv0(x) # (B, 8, H, W)
conv1 = self.conv1(conv0) # (B, 16, H//2, W//2)
conv2 = self.conv2(conv1) # (B, 32, H//4, W//4)
feat2 = self.toplayer(conv2) # (B, 32, H//4, W//4)
feat1 = self._upsample_add(feat2, self.lat1(conv1)) # (B, 32, H//2, W//2)
feat0 = self._upsample_add(feat1, self.lat0(conv0)) # (B, 32, H, W)
# reduce output channels
feat1 = self.smooth1(feat1) # (B, 16, H//2, W//2)
feat0 = self.smooth0(feat0) # (B, 8, H, W)
# feats = {"level_0": feat0,
# "level_1": feat1,
# "level_2": feat2}
return [feat2, feat1, feat0] # coarser to finer features
|