Confusion in the way descriptors of the keypoints are extracted

[rish-av]

Hi,

Thank you for the pytorch implementation of SuperPoint.
Reference code:

def desc_to_sparseDesc(self):
        # pts_nms_batch = [self.getPtsFromHeatmap(h) for h in heatmap_np]
        desc_sparse_batch = [self.sample_desc_from_points(self.outs['desc'], pts) for pts in self.pts_nms_batch]
        self.desc_sparse_batch = desc_sparse_batch
        return desc_sparse_batch

in the file Val_model_heatmap.py, I see that you are using the function desc_to_sparseDesc to get the descriptors of the relevant keypoints, but pts seems to be taken from a heatmap of size (H,W) but outs['desc'] here is (D,H/8,W/8) and there is no upsample operation being done here. How is this getting the descriptors of the relevant heatmaps?

Thanks

[rish-av]

This is simple, nevermind.