Add model details
Browse files
README.md
CHANGED
|
@@ -47,17 +47,43 @@ An image-to-image model for upscaling blurry MNIST-like images.
|
|
| 47 |
```
|
| 48 |
|
| 49 |
## Model Details
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
|
| 51 |
## Training Procedure
|
| 52 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
## Evaluation
|
| 54 |
|
| 55 |
## Ethical Considerations & Limitations
|
|
|
|
| 56 |
|
| 57 |
## Caveats and Recommendations
|
| 58 |
|
| 59 |
## Citation
|
|
|
|
|
|
|
|
|
|
| 60 |
|
| 61 |
## License
|
| 62 |
|
| 63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
```
|
| 48 |
|
| 49 |
## Model Details
|
| 50 |
+
- **Developers:** Blaise Gauvin St-Denis (bstdenis)
|
| 51 |
+
- **Organization:** Ouranos
|
| 52 |
+
- **Version:** v1.0 (2025-08-13)
|
| 53 |
+
- **Dataset:** `mnist` (trainsplit)
|
| 54 |
+
- **Evaluation dataset:** `mnist` (test split)
|
| 55 |
+
- **Compute:** 1x L40 (ToDo training time)
|
| 56 |
+
- **Framework:** Dense UNet / PyTorch 2.8
|
| 57 |
+
- **Hardware/Software dependencies:** CUDA 12.8, Python 3.12
|
| 58 |
|
| 59 |
## Training Procedure
|
| 60 |
|
| 61 |
+
### Data preprocessing
|
| 62 |
+
- Resize to 32 x 32 with all possible locations for the 28 x 28 original data
|
| 63 |
+
- Downcale 4x (to 8 x 8)
|
| 64 |
+
|
| 65 |
+
### Hyperparameters
|
| 66 |
+
- Epochs:
|
| 67 |
+
- Batch size:
|
| 68 |
+
- Learnings rate:
|
| 69 |
+
- Weight decay:
|
| 70 |
+
- Seed: 0
|
| 71 |
+
|
| 72 |
## Evaluation
|
| 73 |
|
| 74 |
## Ethical Considerations & Limitations
|
| 75 |
+
- For educational purpose only
|
| 76 |
|
| 77 |
## Caveats and Recommendations
|
| 78 |
|
| 79 |
## Citation
|
| 80 |
+
If you use this model, please cite
|
| 81 |
+
|
| 82 |
+
ToDo
|
| 83 |
|
| 84 |
## License
|
| 85 |
|
| 86 |
+
This model is released under the Apache 2.0 License. See `LICENSE` file.
|
| 87 |
+
|
| 88 |
+
## Contact
|
| 89 |
+
- **Maintainer:** @bstdenis
|