Add README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
tags:
|
| 4 |
+
- image-denoising
|
| 5 |
+
- onnx
|
| 6 |
+
- nafnet
|
| 7 |
+
pipeline_tag: image-to-image
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# Obscura models
|
| 11 |
+
|
| 12 |
+
ONNX models downloaded on demand by [Obscura](https://obscura.tomfl.dev).
|
| 13 |
+
|
| 14 |
+
## NAFNet-SIDD-width32 (AI denoise)
|
| 15 |
+
|
| 16 |
+
| File | Precision | Size | SHA-256 |
|
| 17 |
+
|---|---|---|---|
|
| 18 |
+
| `nafnet-sidd-w32.onnx` | FP32 | 117,085,438 B | `092dc37894ee36f3ceef1359f73b0e566a24c9194138458cfd92b56d8f74647a` |
|
| 19 |
+
| `nafnet-sidd-w32-fp16.onnx` | FP16 weights and maths, FP32 in/out | 58,957,563 B | `74bb136d6e78ebab8490c7db29b5bda37c270bd382a03990c66d9cc17271f581` |
|
| 20 |
+
|
| 21 |
+
- **Model**: NAFNet, "Simple Baselines for Image Restoration" (Chen, Chu,
|
| 22 |
+
Zhang, Sun — ECCV 2022), https://github.com/megvii-research/NAFNet —
|
| 23 |
+
MIT License, Copyright (c) 2022 megvii-model.
|
| 24 |
+
- **Weights**: Megvii's official `NAFNet-SIDD-width32.pth` (trained on SIDD),
|
| 25 |
+
SHA-256 `89c70e808d1783b6c07911306e106aaf0d4f7f3da8c61078b99ff7f8929a26f4`,
|
| 26 |
+
converted unchanged.
|
| 27 |
+
- **Export**: `tools/ml/export_nafnet.py` in the Obscura repository (ONNX
|
| 28 |
+
opset 17; FP16 by onnxconverter-common with FP32 inputs/outputs).
|
| 29 |
+
- **Input / output**: `input` → `output`, float32 NCHW RGB in 0..1, any
|
| 30 |
+
height/width that are multiples of 16. The network's own padding is left
|
| 31 |
+
out: the caller tiles.
|
| 32 |
+
|
| 33 |
+
MIT License text: https://github.com/megvii-research/NAFNet/blob/main/LICENSE
|