search

IceClear / CLIP-IQA

[AAAI 2023] Exploring CLIP for Assessing the Look and Feel of Images
Other
283 stars 17 forks source link

Image transformation methods #10

Open luutn2002 opened 1 year ago

luutn2002 commented 1 year ago

Hi there, thank you for publishing such a great work. May I ask what is the resolution for CLIP-IQA? The resolution I found in coopclipiqa.py file is 224x224, but the resize transformation in config files are all commented out. So does this mean that there are some code for image resizing somewhere else or the image resolution is kept the same? The resolution are also not included in the paper as well.

IceClear commented 1 year ago

image

luutn2002 commented 1 year ago

Well this part still does not clarify which type of image transformation being used in the code and only mentioned the train and test splitting. Are you suggesting that the data transformation methods are also the same as the paper being mentioned?

IceClear commented 1 year ago

Well this part still does not clarify which type of image transformation being used in the code and only mentioned the train and test splitting. Are you suggesting that the data transformation methods are also the same as the paper being mentioned?

Because no additional image transformation is used unless clarified. clip-iqa does not rely on fixed resolution.