I noticed that there are many pictures in DIV2K and Flickr2k in the validation set and test set, and the corresponding GTs are available. In order to ensure fair competition, please ask if there is a judging method to prevent some participants from using these extra data.
Posted by: weller1212 @ March 6, 2024, 2 a.m.We have noticed feedback from participants indicating the potential for overfitting on the test datasets. We have also observed some submissions attempting to improve metrics through overfitting on the test datasets.
In order to ensure fairness in the competition, we have decided to exclude public datasets such as div2k from the final testing rankings and instead utilize remaining non-public datasets for the final ranking.
For the final ranking, we will refer to the model submitted in the final factsheet. Based on this model, we will conduct rankings and announce the final scores.
Posted by: wuyaqi16 @ March 6, 2024, 6:47 a.m.