abnormal inference result

[Lzars]

I changed my own dataset, with a total of 700 pairs. After training 600 epoch, the F-score is 0.4 and the CDL2 is 2.0, but the points in the inference results are concentrated in the middle of the patch. Many points are crowded near the center point. Why? help

[Eva-xx]

Have you found the reason or a solution to the problem? I also encountered this issue when experimenting with my own dataset — the dense points generated by FoldingNet are clustered around the sparse center points predicted by the Transformer.

[Lzars]

No, I didn't find the reason in the end. do you have any solution> Have you found the reason or a solution to the problem? I also encountered this issue when experimenting with my own dataset — the dense points generated by FoldingNet are clustered around the sparse center points predicted by the Transformer.