Method _aggregate_feature

script/feature/model.py:400–413

assume target and nerf rgb are inferenced at the same time, slice target batch and nerf batch and aggregate features :param x: image blob (2B x C x H x W) :param upsampleH: New H :param upsampleW: New W :return feature: (2 x B x H x W)

(self, x, upsampleH, upsampleW)

        self.fc_pose = nn.Linear(1280, feat_dim)
    
    def _aggregate_feature(self, x, upsampleH, upsampleW):
        '''
        assume target and nerf rgb are inferenced at the same time,
        slice target batch and nerf batch and aggregate features
        :param x: image blob (2B x C x H x W)
        :param upsampleH: New H
        :param upsampleW: New W
        :return feature: (2 x B x H x W)
        '''
        batch = x.shape[0] # should be target batch_size + rgb batch_size
        feature_t = torch.mean(torch.nn.UpsamplingBilinear2d(size=(upsampleH, upsampleW))(x[:batch//2]), dim=1)
        feature_r = torch.mean(torch.nn.UpsamplingBilinear2d(size=(upsampleH, upsampleW))(x[batch//2:]), dim=1)
        feature = torch.stack([feature_t, feature_r])
        return feature

    def _aggregate_feature2(self, x):
        '''