Excuse me. When I reproduce the work to train on the provided preprocessed data of BTCV dataset. I didn't make any other changes except 1)using a single GPU (RTX 4090) instead; 2) using pre-trained SAM ViT_B. However, NAN appears at about epoch 4/400.
I insert some code snippets to help find the reason, through which I find the origin of NAN occurs no later than the output['low_res_logits'] .
Could you please offer some hints to further identify and resolve the problem?
Excuse me. When I reproduce the work to train on the provided preprocessed data of BTCV dataset. I didn't make any other changes except 1)using a single GPU (RTX 4090) instead; 2) using pre-trained SAM ViT_B. However, NAN appears at about epoch 4/400.
I insert some code snippets to help find the reason, through which I find the origin of NAN occurs no later than the output['low_res_logits'] .
Could you please offer some hints to further identify and resolve the problem?