CLIP-IQA
CLIP-IQA copied to clipboard
About dataset setting.
Hi! I am just wondering about the CLIPIQA+ experiment mentioned in the paper. In Table 1, it says that the model was trained on Koniq and tested on Koniq, LIVE-itW, and SPAQ. So, I was curious about the specifics of the training process. Did it involve training on the official Koniq-train dataset, selecting the best parameters on Koniq-valid, and testing on Koniq-test, LIVE-itW, and SPAQ?
Hi. We did not use Koniq-valid. We just directly used the latest ckpt for test.