Find that FPS random start sampling results in different CD at 1024 points

[leonardozcm]

                # downsample gt to 2048
                partial = fps_subsample(gt, 2048)
                coarse_gt = fps_subsample(gt, 1024)

                # preprocess transpose
                partial = partial.permute(0, 2, 1)

                v, y_coarse, y_detail = model(partial)

                # y_coarse = y_coarse.permute(0, 2, 1)
                y_coarse = fps_subsample(
                    gt[:, torch.randperm(gt.shape[1]), :], 1024)
                y_detail = y_detail.permute(0, 2, 1)

                loss_coarse = chamfer_sqrt(coarse_gt, y_coarse)
                loss_fine = chamfer_sqrt(
                    gt, gt[:, torch.randperm(gt.shape[1]), :])
                loss = loss_coarse + 0.1 * loss_fine

============================ TEST RESULTS ============================
Taxonomy #Sample ChamferDistance
02691156 10 0.0000
02933112 9 0.0000
02958343 10 0.0000
03001627 9 0.0000
03636649 9 0.0000
04256520 10 0.0000
04379243 9 0.0000
04530566 9 0.0000
Overall 0.0000

Epoch 11 11.2085 0.0000 11.2085