def forward()

in models/base_model.py [0:0]

• 24 lines of code
• 12 McCabe index (conditional complexity)

    def forward(self, video, *args, **kwargs):
        """
            Args: video (torch.Tensor)
                Could be (B, #clips, C, T, H, W) or
                    (B, #clips, #crops, C, T, H, W)
            Returns:
                Final features
                And any auxiliarly losses produced by the model
        """
        if video.ndim == 6:
            video_crops = [video]
        elif video.ndim == 7 and video.size(2) == 1:
            video_crops = [video.squeeze(2)]
        elif video.ndim == 7:
            video_crops = torch.unbind(video, dim=2)
        else:
            raise NotImplementedError('Unsupported size %s' % video.shape)
        feats_losses = [
            self.forward_singlecrop(el, *args, **kwargs) for el in video_crops
        ]
        feats, losses = zip(*feats_losses)
        # Convert to dict of lists
        feats = {k: [dic[k] for dic in feats] for k in feats[0]}
        losses = {k: [dic[k] for dic in losses] for k in losses[0]}
        # Average over the crops
        feats = {
            k: torch.mean(torch.stack(el, dim=0), dim=0)
            for k, el in feats.items()
        }
        losses = {
            k: torch.mean(torch.stack(el, dim=0), dim=0)
            for k, el in losses.items()
        }
        return feats, losses