Open grtzsohalf opened 2 months ago
Hi, here are some suggestions:
Thanks for your suggestion! I found the issue arose from the overfitting. After making the network smaller, the validation results seem reasonable.
Thanks for your suggestion! I found the issue arose from the overfitting. After making the network smaller, the validation results seem reasonable.
That does sound work. Usually, this is usually caused by the small-scale training assets, and restricting model capacity is a choice to solve it.
Hi, I am training a model on my custom data using the following config
where
MyDataset
is modified a little bit fromDefaultDataset
(my data doesn't contain normals)My data only contains coord, color, scene_id, and semantic_gt.
The training and validation results look good in several epochs initially, but the validation metrics suddenly drop around epoch 23. Is it simply an overfitting? Do you have any guess about it? Thanks!