README.md

---
license: mit
pipeline_tag: image-segmentation
library_name: pytorch
tags:
  - unet
  - regnetz_d8
  - segmentation-models-pytorch
  - timm
  - pytorch
  - remote-sensing
  - sentinel-2
  - multispectral
  - cloud-detection
datasets:
  - isp-uv-es/CloudSEN12Plus
---

# Cloud Detection — U-Net (RegNetZ D8 encoder)

**Repository:** `Burdenthrive/cloud-detection-unet-regnetzd8`  
**Task:** Multiclass image segmentation (4 classes) on **multispectral Sentinel‑2 L1C** (13 bands) using **U‑Net** (`segmentation_models_pytorch`) with **RegNetZ D8** encoder.

This model predicts per‑pixel labels among: **clear**, **thick cloud**, **thin cloud**, **cloud shadow**.

---

## ✨ Highlights

- **Input:** 13‑band Sentinel‑2 L1C tiles/patches (float32, shape `B×13×512×512`).  
- **Backbone:** `tu-regnetz_d8` (TIMM encoder via `segmentation_models_pytorch`).  
- **Output:** Logits `B×4×512×512` (apply softmax + argmax).  
- **Files:** `model.py`, `config.json`, and weights.

---

## 📦 Files

- `model.py` — defines the `UNet` class (wrapper around `smp.Unet`).
- `config.json` — hyperparameters and class names:
  ```json
  {
    "task": "image-segmentation",
    "model_name": "unet-regnetz-d8",
    "model_kwargs": { "in_channels": 13, "num_classes": 4 },
    "classes": ["clear", "thick cloud", "thin cloud", "cloud shadow"]
  }