problems during training

[sharonlee12]

hello！I trained on the LITS dataset,and during training,dice loss and bce loss can be reduced during training, but during the test, the results are all 0: (epoch 500): Liver: dice 0.0000, recall 0.0000, precision 0.0000. Liver Tumor: dice 0.0000, recall 0.0000, precision nan. case04_LiTS/label/liver_1| Liver: 0.0000, Liver Tumor: 0.0000, Case04_LITS /label/liver_1| liver: 0.0000, liver tumor: 0.0000, At the beginning (epoch 50), the result is Liver: dice 0.1749, recall 0.4666, precision 0.1077. Liver Tumor: dice 0.0032, recall 1.0000, precision 0.0016. I am confused about this, what is the possible reason? I look forward to your reply, thank you!

[sharonlee12]

In addition, the results with the pre-trained model you provided are normal： Liver: dice 0.9786, recall 0.9776, precision 0.9796. Liver Tumor: dice 0.8918, recall 0.9217, precision 0.8640. case04_LiTS/label/liver_10| Liver: 0.9786, Liver Tumor: 0.8918,

[Adoreeeeee]

I'm having the same issue as you. I think it could be that the model is too big and needs enough data to train. A single small data set cannot satisfy parameter learning.

[ljwztc]

The convergence of this model may require a substantial dataset, although it is not necessarily the sole contributing factor.