Closed jiangxb98 closed 8 months ago
The difference is reasonable, given the mini_trainval is randomly generated as here. The chamfer_distance_inner key is removed. It is now updated as the chamfer_distance key.
I don't see any explicit random generation, is the random generation due to this code os.listdir('dataset/openscene-v1.1/meta_datas/mini')
?
Yes.
Thanks.
When I used 1/8 of the mini data, the training data was 5108, but the github log printed is 621 (621*8=4968), and the val data length also not match (5554 vs 6462).
In addition, the evaluation indicators are also different, and the chamfer_distance_inner indicator is missing.