Yanze Wu

The leftmost column of the table is not actually the original inputs, they are the **resized cropped patches** from the original inputs for the ease of comparison. You can get...

Hi, we will sonn release a new version of animevideo model (RealESRGAN-animevideo**v3**-x4) which is more faithful to the original colors. Please stay tuned. 

Hi, we will soon release a new version of animevideo model (RealESRGAN-animevideov3-x4), which is optimized for the restoration of artificially blurred scenes. Please stay tuned. 

Hi, you can find these models in [Google Drive Link](https://drive.google.com/drive/folders/1BYjmwiiOTiBrnqW9TuYBpIXFL1X-JWVc).

ImageNet is used for training and testing, as described in Section 4.1&4.2 of the paper.

Currently, we don't have a plan to release the training code. I think the critical implementation details are already available from the paper, supplementary material, and this repo. Can you...

Sorry for the late reply. I just added the codes&checkpoint&examples for diverse colorization. Hope it helps~

Recently, I am busy on other projects and don't have enough time to prepare&release the training code. However, since the training details are already described in the paper&supp, so you...

Hi, if you are testing in-the-wild images, this AssertionError indicates that `assets/predicted_label_for_user_image.txt ` file doesn't contain the predicted label for your testing images. Please check the instruction [here](https://github.com/ToTheBeginning/GCP-Colorization#test-in-the-wild-images).

This looks like a timm problem, maybe you can check if timm (my timm version is 0.6.5) is installed successfully, for example, check if `beit_large_patch16_512` is in `timm.list_models()`.