Closed Atmyre closed 9 months ago
The enhancement model I used to report in my paper is pretrained_models/enhancement_model.pth.
Thank you for your reply
But I tried to run test.py on this model, and I get quite bad results. Like this:
I run the code with this command: python test.py -i ./test_data/BAID/test/resize_input/ -o ./inference_results/ -c ./pretrained_models/enhancement_model.pth
I guess this does not look like pictures from the paper. Could you suggest what I could possibly be doing wrong?
And the metrics I get on BAID are the following: PSNR: 9.51 SSIM: 0.61 LPIPS: 0.34
I check the test data, the checkpoint and the code on this github repo once more. And only find that the test data has some duplicate images in the BAID test dataset I provided through Google Drive link (I apologize for not noticing this earlier. I will delete the duplicate ones soon).
I download the code, the checkpoint and the BAID test data from this repo again and rerun the test experiments.
For the image you mentioned, I think it is BAID test dataset 0060.JPG. The result I have just generated for 0060.JPG is:
The metrics I have just got on BAID test dataset are: PSNR: 21.611 SSIM: 0.883 LPIPS: 0.159 MUSIQ: 55.683 Even a little bit better than what I repoerted in the paper. The metrics may be changed a little due to different metrics codes, but could never be something like what you mentioned above.
The command you use is correct. Maybe you changed the input test data or the code?
Hello,
Sorry for the late reply, just seen it not so long ago. Thanks for checking, there was a mistake it my code indeed. Sorry for bothering
I also wanted to ask if you could provide weights for the best trained model, results for which you have reported in the paper?