Question regards .57 when computing F1 score.

[zshyang]

Dear Edward Smith,

The code to compute f1 score
dist_to_pred = torch.sqrt(torch.sum((.57*pred_counters - .57*gt_points.contiguous().view(-1,3))**2, dim = 1 )).view(batch_size, -1) https://github.com/EdwardSmith1884/GEOMetrics/blob/09471d70f441482319543bb5214eaaab38f64507/utils.py#L490

I am not sure why .57 is needed?

[EdwardSmith1884]

This is the scaling everyone does for the 3Dr2n2 Dataset. When they made the images they scaled down the objects by this amount, so I do the same to make the object have the same size. You can find examples of this is other repos too.

[zshyang]

Thanks a lot!