Upload 3 files

README.md

@@ -0,0 +1,44 @@
+# EDSR-SR-DSC (4× Super-Resolution for Wind Data)
+
+This model is a custom-trained version of the Enhanced Deep Super-Resolution (EDSR) model from the [`super-image`](https://github.com/eugenesiow/super-image) library. It is adapted for super-resolution of **2-channel weather data** (e.g., wind u and v components), upscaling coarse-resolution wind fields by a factor of 4×.
+
+---
+
+## 🧠 Model Architecture
+
+- **Base**: EDSR ([Lim et al. 2017](https://arxiv.org/abs/1707.02921))
+- **Input channels**: 2 (not RGB)
+- **Output channels**: 2
+- **Feature channels (`n_feats`)**: 64
+- **Residual blocks**: 32
+- **Upsampling**: Enabled
+- **Scale factor**: 4×
+
+Mean-shift normalization layers were removed (`sub_mean`, `add_mean`), as the model was trained on standardized wind data.
+
+---
+
+## 📦 Files in this Repository
+
+| File                   | Description                                   |
+|------------------------|-----------------------------------------------|
+| `config.json`          | Configuration compatible with `super-image`   |
+| `pytorch_model_4x.pt`  | Pretrained model weights (4× model)           |
+
+---
+
+## 🚀 How to Use
+
+```python
+from super_image import EdsrModel, ImageLoader
+import torch
+
+# Load model
+model = EdsrModel.from_pretrained("lschmidt/edsr-sr-dsc", scale=4)
+
+# Load a 2-channel tensor (e.g., from a NetCDF or preprocessed image)
+# Example assumes shape (1, 2, H, W)
+inputs = torch.randn(1, 2, 64, 64)  # replace with your real data
+
+# Super-resolve
+outputs = model(inputs)

config.json

@@ -0,0 +1,11 @@
+{
+  "model_type": "EDSR",
+  "n_colors": 2,
+  "n_feats": 64,
+  "n_resblocks": 32,
+  "no_upsampling": false,
+  "res_scale": 1,
+  "rgb_range": 255,
+  "scale": 4
+}
+

edsr_dsc_weather_4x.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60a0798fdd2b001ce82b3065b25e08d8179b346e96cef287c2129107cdb28d51
+size 6093322