Where can I obtain the 4x-DeCompress model for artifact removal?

[Zhubisong]

In the JPEG Artifact Correction module on your website, I found that the 4x-DeCompress model as a whole seems to have better effects. However, I didn't find the weights of this model in your repository. May I ask where I can find it?

[Phhofm]

Hey
That model was not trained by me but by another model trainer, this is why you wont find it in my repository of course.
The model you are referring to was trained by @Kim2091
You would find this model in the model-releases channel on the Enhance Everthing discord server.

For simplicity, i paste the text of the release here for you. The pcloud link is still working and contains the model files you are looking for

<@&560103931204861954> <@&577839492199874570>
Name: 4x-DeCompress and Decompress-Strong
License: CC BY-NC-SA 4.0
Link: https://u.pcloud.link/publink/show?code=kZL4XhVZB8qmB8uXcahRGnGRMIyC0JyaQoxy https://mega.nz/folder/vBgGgA5Z#qXsikSiIU8Edemaju91zDA
Model Architecture: ESRGAN
Scale: 4
Purpose: This is UniScaleV2, but intended for images with compression. Seems to work best on realistic images.

Iterations: ~ 130k
batch_size: 4
HR_size: 112
Epoch: 23-28
Dataset: Custom Dataset consisting of lossless 4k frames from Metal Arms: Glitch in the System, Just Cause 3, Dirt 3, Forza Horizon 3, Sleeping Dogs, and self-edited photos from SignatureEdits + ATLA DVD images
Dataset_size: 18,909 tiles + 288 ATLA Frames
OTF Training Yes (JPEG artifacts, camera, poisson base_blur, bsrgan_resize)
Pretrained_Model_G: 4xESRGAN

Description: This model set is UniScaleV2, but trained to remove image compression while bringing out details. In my tests, it comes close to DF2K_JPEG in many cases, and provides more realistic results in others. Hope you guys enjoy 🙂

The Strong model is meant for images with very strong JPEG compression. It'll work on normal or basic compressed images, but it'll lose detail.

There's an Extreme model included, but it can introduce a LOT of artifacts. Only use it if the others aren't working.

Seems to work best on realistic images.

Comparisons:
https://cdn.discordapp.com/attachments/884864597310459996/884952891721383936/unknown.png
https://cdn.discordapp.com/attachments/884864597310459996/884957054807179274/unknown.png
https://cdn.discordapp.com/attachments/884864597310459996/884960697883193354/unknown.png
https://cdn.discordapp.com/attachments/884864597310459996/884953810097827850/unknown.png

[Zhubisong]

Thank you very much!

[Zhubisong]

I have another question.

1. Nowadays, the AI performance of mobile phones has become very powerful, and many super-resolution models can run in real-time on mobile devices. However, these super-resolution models are all classic ones, which means they can only process clean, low-resolution images. But if an image contains multiple degradations such as blocking effects, compression artifacts, noise, etc., these models cannot handle them. I would like to ask: Is there any highly efficient model that can handle multiple degradations and run on mobile phones?
2. In addition, I would like to know whether you have paid attention to video super-resolution/restoration in the above-mentioned complex scenarios?
   Thanks again!