mlmPenguin commited on
Commit
538c0d8
·
verified ·
1 Parent(s): e0b8640

Add model card

Browse files
Files changed (1) hide show
  1. README.md +91 -0
README.md ADDED
@@ -0,0 +1,91 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ tags:
4
+ - image-super-resolution
5
+ - image-to-image
6
+ - mlx
7
+ - apple-silicon
8
+ - super-resolution
9
+ - neural-heat-fields
10
+ base_model:
11
+ - prs-eth/thera-rdn-air
12
+ - prs-eth/thera-rdn-pro
13
+ library_name: mlx
14
+ language:
15
+ - en
16
+ pipeline_tag: image-to-image
17
+ ---
18
+
19
+ <div align="center">
20
+
21
+ # ✨ thera-mlx
22
+
23
+ **Aliasing-Free Arbitrary-Scale Super-Resolution for Apple Silicon**
24
+
25
+ *An [MLX](https://github.com/ml-explore/mlx) port of [Thera](https://therasr.github.io) — running natively on M1/M2/M3/M4*
26
+
27
+ [![GitHub](https://img.shields.io/badge/GitHub-madeleinelmuller%2Fthera--mlx-black?logo=github)](https://github.com/madeleinelmuller/thera-mlx)
28
+ [![License](https://img.shields.io/badge/License-Apache%202.0-blue)](https://github.com/madeleinelmuller/thera-mlx/blob/main/LICENSE)
29
+ [![Paper](https://img.shields.io/badge/Paper-arXiv%3A2311.17643-red)](https://arxiv.org/abs/2311.17643)
30
+
31
+ **128×128 → 512×512 in ~1.5 seconds on M1** · Arbitrary scale factors · Web UI + CLI · Video support
32
+
33
+ </div>
34
+
35
+ ---
36
+
37
+ ## Model Description
38
+
39
+ **thera-mlx** contains MLX-format weights for the [Thera](https://therasr.github.io) super-resolution model, converted from the original JAX/Flax checkpoints released by [PRS-ETH](https://prs.igp.ethz.ch/) (ETH Zurich).
40
+
41
+ Thera uses **neural heat fields** to perform aliasing-free image super-resolution at *any* scale factor — not just 2× or 4×. This repo provides two variants:
42
+
43
+ | File | Model | Speed | Quality |
44
+ |------|-------|-------|---------|
45
+ | `weights-air.safetensors` | Air | ~1.5s for 128→512px on M1 | Great |
46
+ | `weights-pro.safetensors` | Pro (+ SwinIR refinement) | ~5s for 128→512px on M1 | Best |
47
+
48
+ ## Usage
49
+
50
+ ```bash
51
+ git clone https://github.com/madeleinelmuller/thera-mlx
52
+ cd thera-mlx
53
+ pip install -r requirements.txt
54
+
55
+ # Download weights (pulls from this repo automatically)
56
+ python run.py convert --model air
57
+ python run.py convert --model pro
58
+
59
+ # Launch web UI
60
+ python run.py
61
+
62
+ # Or use CLI
63
+ python run.py run input.png output.png --scale 4 --model air
64
+ ```
65
+
66
+ See the full [GitHub repository](https://github.com/madeleinelmuller/thera-mlx) for complete documentation.
67
+
68
+ ## Original Weights
69
+
70
+ The weights in this repository are converted from:
71
+ - [`prs-eth/thera-rdn-air`](https://huggingface.co/prs-eth/thera-rdn-air)
72
+ - [`prs-eth/thera-rdn-pro`](https://huggingface.co/prs-eth/thera-rdn-pro)
73
+
74
+ ## Citation
75
+
76
+ If you use this in your work, please cite the original Thera paper:
77
+
78
+ ```bibtex
79
+ @article{becker2023thera,
80
+ title = {Thera: Aliasing-Free Arbitrary-Scale Super-Resolution with Neural Heat Fields},
81
+ author = {Becker, Alexander and Caye Daudt, Rodrigo and Narnhofer, Dominik and
82
+ Peters, Torben and Metzger, Nando and Wegner, Jan Dirk and Schindler, Konrad},
83
+ journal = {arXiv preprint arXiv:2311.17643},
84
+ year = {2023}
85
+ }
86
+ ```
87
+
88
+ **Original resources:**
89
+ - Paper: [arxiv.org/abs/2311.17643](https://arxiv.org/abs/2311.17643)
90
+ - Project page: [therasr.github.io](https://therasr.github.io)
91
+ - Original HF weights: [huggingface.co/prs-eth](https://huggingface.co/prs-eth)