-
Notifications
You must be signed in to change notification settings - Fork 0
/
notes.txt
37 lines (30 loc) · 4.12 KB
/
notes.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
- Throw all images from the public folder to mozjpeg to compress, if already compressed nothing will happen. Upload. Make sure they are all compressed. https://mozjpeg.com/
- Then check in the anime examples if image size matters for your vue comparison component. If not you can leave them original size, otherwise resize all to 2x since 2x and 4x outputs are used.
- Check examples maybe extend them with 2x examples if works, i did not include them at fist because 2x against 4x comparison seemed unfair but i mean it can still be visually compared.
- Anime examples resize/downsize to 2x since its 2x models and 4x models etc mixed. Then upload images. And check examples on multimodel page again. PS dont forget to move sources folder into the docs/public for it to be loaded into dist, remove sources folder outside afterwards to avoid confusion.
- Update Favorites page (like the hat real gan model and maybe others you had added recently since the last creation of the fav page)
- Add more categories like JPEG Artifacts or halo removal, demoire etc
- Create input images for the JPEG Artifacts examples. Then create examples.
- Add more models like ART (Accurate Image Restoration with Attention Retractable Transformer) use the google colab i made for this (only adde the code parts for upscaling so far, but can also be used for denoise)
- Also use / add models that you can find in the discord model relases category of that upscale wiki model discord. A lot of them have not been added to the wiki yet I believe
- Maybe help adding those models you find which have not been added to the wiki yet? Cant see bubbles animesscale sinir model when i use the text search hmm
Ah maybe people should specifically request to have their model put on there because of licencing stuff? Not sure. Im pretty sure everyone that actually publishes his model on the model releases page would also have it on the wiki, i mean, they already published their model. If they only show images as examples but dont publish their model then thats a different story but then i cant also add it so that works.
To be honest, it was a little bit a pain getting multiple models from this page like the anime ones, since multiple of them used 1fichier or something which had a 30 min time limit after getting one model to download the next ... tor browser to the rescue with establishing new connections ... a pain
- Make a Upscale Wiki Models page, where I only use the models that are exactly listed on the upscale wiki per category, so an example (or multiple) per category. I mean this is how this whole page started basically. But I added/used more and more models (from the replicate, huggingface or simply google colabs or own written colabs) that are not from that site and not supported by chaiNNer. Would be cool if i make a single page with only models from the wiki so we know most of them can be used with chaiNNer.
- Replace all imgsli examples with own component at some point so update is easy (just manage files)
- Include a Section for replicate models and huggingface models? Like models you can use over those sites?
I think I have not added FeMaSR output results to multimodel current yet? check
Categories:
- JPEG Artifacts Removal
- Demoire
- Dehalo
- Colorize (Black and White images)
Train your own model? You will learn more this way about how all of this works (experience)
---------------------------------
Keep an eye out for new models you want to test
https://github.com/bahjat-kawar/ddrm-jpeg JPEG Artifact Correction using Denoising Diffusion Restoration Models
https://github.com/gladzhang/art Accurate Image Restoration with Attention Retractable Transformer
https://huggingface.co/spaces/danielsapit/JPEG_Artifacts_Removal Accurate Image Restoration with Attention Retractable Transformer https://huggingface.co/spaces/danielsapit/JPEG_Artifacts_Removal
https://github.com/wyhuai/ddnm Zero-Shot Image Restoration Using Denoising Diffusion Null-Space Model
https://github.com/icandle/MAN Multi-scale Attention Network for Single Image Super-Resolution
https://github.com/csxmli2016/redegnet From Face to Natural Image: Learning Real Degradation for Blind Image Super-Resolution