executing Execute the training file, but it has not been able to converge, and the prediction result of the verification data set is wrong.

[zxm115]

 Excuse me , I want to ask you a question about executing the training file named train_aaf.py  using PASCAL VOC dataset, which it has not been able to converge, despite following the default parameter settings consistent with the paper. And the prediction result of the validation data set is wrong, sometimes it can display black images with noise points, but in most cases  it shows all black, I suspect that there is a problem with model training. However when I download the trained models which you offer, the results based on the test data set are consistent with the paper. Could you please give me some advice about this situation, thanks so much. If you could offer more detailed parameter setting instructions, I would be grateful.

[twke18]

Hi, please check our bashscripts examples. The parameters listed in the bash files are the ones we reported in the paper.

For training, perhaps you can first try training the baseline models for debugging.

[zxm115]

Thank you very much for your help. Can you help me see this image. Why is the prediction result of the validation data set above always showing no content?

[wanghaokevin]

Thank you very much for your help. Can you help me see this image. Why is the prediction result of the validation data set above always showing no content?

Have you solved this problem?I am training the baseline with the parameters of the bashscipts, but there is also no concept .