about evaluation about ddcolor HOT 2 CLOSED

The answer is no. For validation, we follow the same approach as previous methods, i.e. for ImageNet val50k, COCO-Stuff, and ADE20K, the test images are resized into 256 x 256 (same as ColorFormer); for ImageNet val5k, the test images are center cropped and resized into 256 x 256 (same as CT2).

from ddcolor.

Thank you for your interest in our research and for raising your concerns.

1. Yes, val5k means the first 5000 images in validation set.
2. For the △colorful, we calculated the average cf of ImageNet val5k to be 38.21. So we actually got better results with CT2 than what they reported in their paper.
3. The results described in our paper were obtained experimentally. However, we acknowledge that real-world results may vary due to various factors such as differences in machines and running environments. To better understand the differences, we encourage you to replicate our experiments in your own environment.

While we appreciate the opportunity to discuss further, please understand that our time and resources are limited and we need to focus on our ongoing research. Thank you!

from ddcolor.