Unfair comparison
xiaobiaodu opened this issue · 2 comments
Your IAT is trained on 689 LOL training images.
But the original LOL dataset just contains 485 training images.
Table 1 shows the results of the LOL dataset. Other methods, like MAXIM, are trained on the original dataset.
More training data means better results. It seems the author extends external training data for better results.
The author should give a reasonable explanation for why conducts so unfair comparison.
Really thanks for your attention, we have renew our Arxiv version, and the renewed results on LOL-V1 results is 23.38 and 0.809. The new Arxiv version will be update this week. I'll release the training code and test code on LOL-V1 dataset as soon as possible. Any question else please drop here~
Good job. Thanks for your reply!