| 435 | x = self.prob(x) |
| 436 | return x |
| 437 | class AggWeightNetVolume(nn.Module): |
| 438 | def __init__(self, in_channels=32,hid_channels=1,out_channels=1,relu=True): |
| 439 | super(AggWeightNetVolume, self).__init__() |
| 440 | self.w_net = nn.Sequential( |
| 441 | Conv3d(in_channels, hid_channels, kernel_size=1, stride=1, padding=0,relu=relu), |
| 442 | Conv3d(hid_channels, out_channels, kernel_size=1, stride=1, padding=0,relu=relu) |
| 443 | ) |
| 444 | |
| 445 | def forward(self, x): |
| 446 | """ |
| 447 | :param x: (b, c, d, h, w) |
| 448 | :return: (b, 1, d, h, w) |
| 449 | """ |
| 450 | w = self.w_net(x) |
| 451 | return w |
| 452 | |
| 453 | |
| 454 | def depth_regression(p, depth_values,axis=1): |