| 248 | class FrameSimilarity(nn.Module): |
| 249 | |
| 250 | def __init__(self, |
| 251 | in_channels, |
| 252 | inner_channels, |
| 253 | similarity_dim=128, |
| 254 | lookup_window=101, |
| 255 | output_dim=128, |
| 256 | stop_gradient=False, |
| 257 | use_bias=True): |
| 258 | super(FrameSimilarity, self).__init__() |
| 259 | |
| 260 | self.projection = Linear_(in_features=in_channels, out_features=similarity_dim, |
| 261 | bias=use_bias, act="Identity") |
| 262 | self.fc = Linear_(in_features=inner_channels, out_features=output_dim, bias=True, act="ReLU") |
| 263 | |
| 264 | self.lookup_window = lookup_window |
| 265 | self.stop_gradient = stop_gradient |
| 266 | assert lookup_window % 2 == 1, "`lookup_window` must be odd integer" |
| 267 | if torch.cuda.is_available() is True: |
| 268 | self.device = "cuda" |
| 269 | else: |
| 270 | self.device = "cpu" |
| 271 | |
| 272 | def forward(self, inputs): |
| 273 | # pt version [BS, C, N, H, W], so [3, 4] means apply avg on spatial dim, out dim is [BS, C, N] |