Closed vijay-jaisankar closed 6 months ago
Thanks for the advice! We now add a regenerate_crops option for custom images in quality_and_norm_demo.py
You can now use it with "python quality_and_norm_demo.py --regenerate_crops --img car.jpeg" for testing your image.
Thank you for following up on our work. This reply is a supplement to YangiD's reply. The code for cropping and saving images can be found on lines 33-47. For evaluating the quality of your own custom image, please first crop it into _patchnum patches, each measuring 224x224 pixels. Subsequently, save these cropped patches in the directory _./imagesfixedcrop.
We have successfully tested the code with your custom image and the result is as follows.
For car.png, the L_1 norm of output's gradint in term of the input image: HyperIQA:4080.3151 HyperIQA+NT:747.7407 For car.png, the predicted score of the image: HyperIQA:65.0545 HyperIQA+NT:60.2501
If you have any other questions, please feel free to contact us.
Hello, thanks a lot for the prompt fix! Yes, I am now able to use this new config option for custom images. Thanks again for open-sourcing your amazing work.
Hello,
Thank you for open-sourcing your work! While trying to run the model on a custom image, I'm facing an issue, can you please help out with the same?
In particular, the demo (to get the IQA scores) is working on the
123.bmp
image, but not on a custom image. Presumably this is because the cropped patches are present for123.bmp
and not for our custom image. Can you please point out the code to save these cropped patches into the corresponding folder for the custom image?Reproducibility: https://colab.research.google.com/drive/16EL2vuKyIDXjYOCtqxnLCKSKHzGtwN7C?usp=sharing
Thanks!