search

wqb-uestc / DQA

13 stars 3 forks source link

DQA #2

Open feng150416 opened 4 years ago

feng150416 commented 4 years ago

Thanks for your work in the field of image deraining!But I try to train from scratch using the default values of 'test_id.txt',the results are not consistent with the results reported in our paper.I'm so confused,can you tell me why？

wqb-uestc commented 4 years ago

Thanks for your work in the field of image deraining!But I try to train from scratch using the default values of 'test_id.txt',the results are not consistent with the results reported in our paper.I'm so confused,can you tell me why？

The default batch size is 2 in the current code, which enables most users to run our demo with low memory. To reproduce the result in our paper, please find the /DQA/MODEL/BFEN/option.py and change the batch size to 16 as described in our paper. It is noted that the model parameters are randomly initialized in every trial. So the performance will slightly vary across different trials. In our experiment, the performance variation is within +/- 0.02. If you have any other questions, please feel free to contact me.

feng150416 commented 4 years ago

thanks for your reply！