search

xmindflow / deformableLKA

[WACV 2024] Beyond Self-Attention: Deformable Large Kernel Attention for Medical Image Segmentation
https://arxiv.org/abs/2309.00121
157 stars 12 forks source link

Problem encountered during reproduction #9

Closed yangz9527 closed 5 months ago

yangz9527 commented 5 months ago

Hello author, I noticed a strange phenomenon while reproducing the results of DAEFormer on the ISIC2018 dataset. After starting training on the preprocessed dataset, I noticed a gradient explosion in the loss value section. May I ask if this is caused by some issues with the code of the processed dataset 5eaf3fa7038d7c9b93bd80146179c78 a6dbd2a6b0df1868f1c716150a08810

Leonngm commented 5 months ago

Hi, thanks for the interest in our code. The results for the DAEFormer were copied from the DAEFormer publication and not reproduced for this work. Please check if you preprocessed the data with the preprocessing script. Since the loss is already nan after the first iteration, it could also be a different problem.

yangz9527 commented 5 months ago

Hi, thanks for the interest in our code. The results for the DAEFormer were copied from the DAEFormer publication and not reproduced for this work. Please check if you preprocessed the data with the preprocessing script. Since the loss is already nan after the first iteration, it could also be a different problem.

Thank you for taking the time to read and reply to my comments. In fact, I am currently reproducing the results of your previous research on the skin disease ISIC2018 dataset in the DAEformer paper. If possible, could you please update the training and testing codes for the skin disease dataset in the DAEformer? This is very important to me. Thank you very much

Leonngm commented 5 months ago

To train the DAEFormer model on the skin dataset, replace the model in this line with the DAEFormer model.