IceClear/CLIP-IQA

Only return single attribute value for inferencing single image demo

rumiiinator opened this issue · 10 comments

Hi @IceClear,

Thank you for open sourcing this impressive work! I have a question regarding the clipiqa_single_image_demo.py script. When I use this script to do inference on a single image, I notice that only the first value of the attribute list is returned. The output looks something like this: [0.74316406] (using the example rock image).

The print info would be something like:
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
[0.74316406]

Could you please provide some guidance on how to obtain all the attribute values, such as 'Brightness', 'Sharpness', 'Noisiness', 'Colorfulness', and 'Contrast', which may differ from each other?

Thank you very much for your help!

Hi.
Thanks for your interest in our work.
For different attributes, they are actually different attribute pairs.
You only need to change the classname in the config file: here.
I will provide a colab demo recently.

Hi,
Could you provide the pretrained model for attributes values [Brightness', 'Sharpness', 'Noisiness', 'Colorfulness', and 'Contrast']
Thanks

Hi.
We do not need training on assessing these attributes. Just CLIP model is required.

Thanks for your reply.
How can I set the appropriate model setting for attributes values [Brightness', 'Sharpness', 'Noisiness', 'Colorfulness', and 'Contrast']?
I got mismatch error for model and loaded checkpoint

Thanks for your reply. How can I set the appropriate model setting for attributes values [Brightness', 'Sharpness', 'Noisiness', 'Colorfulness', and 'Contrast']? I got mismatch error for model and loaded checkpoint

See here

Hi,
I modify clipiqa_attribute_test.py
and run python demo/clipiqa_single_image_demo.py --config configs/clipiqa/clipiqa_attribute_test.py --checkpoint ./iter_80000.pth
image

and got this error
image

You loaded the wrong model. The model is for CLIPIQA+ (CLIPIQAPredictor). You do not need to load any model for testing CLIPIQA (CLIPIQAFixed).

got it!
Thanks

Hi. Thanks for your interest in our work. For different attributes, they are actually different attribute pairs. You only need to change the classname in the config file: here. I will provide a colab demo recently.

How about the colab demo? where could I find it?
BRs!

Hi. Sry that I am busy with other works so far. May do it when I am available later, possibly after CVPR submission.