Add README with Space metadata1

README.md

@@ -4,8 +4,6 @@ emoji: 🛰️
 colorFrom: blue
 colorTo: gray
 sdk: gradio
-sdk_version: 4.37.2
-python_version: 3.10
 app_file: app.py
 pinned: false
 license: apache-2.0
@@ -18,201 +16,24 @@ An interactive Hugging Face Space for segmenting bridge inspection images into:
 - **beton** (concrete)
 - **steel**
 
-The model architecture is **UNet++**. Large weights are stored separately in a model repo and downloaded automatically on first run (cached afterwards).
+> First launch may take a few minutes while the ~800MB weight file downloads from `g-loubna/bridge-unetpp`.
 
-> First launch may take a few minutes while the ~800MB weight file downloads.
+## Features
+- Colored Mask / Overlay / Raw Class Indices modes
+- Adjustable background transparency (Colored Mask)
+- Adjustable overlay opacity (Overlay)
+- Legend with pixel counts & percentages
+- Downloadable raw mask
 
----
-
-## 🚀 Quick Start (Using the Space)
-
-1. Open the Space page (this UI).
-2. Upload or paste an RGB image (bridge / structural element).
-3. Choose a visualization mode:
-   - Colored Mask
-   - Overlay
-   - Raw Class Indices (grayscale)
-4. (Optional) Adjust:
-   - Overlay Opacity (for Overlay mode)
-   - Background Opacity (for Colored Mask mode)
-5. Click **Run Segmentation**.
-6. Review legend + per‑class pixel counts & percentages.
-7. Download the raw mask (PNG) if needed.
-
----
-
-## 🧠 Model Weights
-
-Weights are pulled from the model repository:
-
-```
-g-loubna/bridge-unetpp
-```
-
-In `app.py`:
-```python
-HUB_REPO_ID = "g-loubna/bridge-unetpp"
-WEIGHTS_FILENAME = "MILESTONE_090_ACHIEVED_iou_0.9077.pth"
-```
-
-Modify these if you update or rename the weight file. Keep the filename stable to leverage caching.
-
----
-
-## 🗂 Repository Structure
-
-| File | Purpose |
-|------|---------|
-| `app.py` | Gradio interface + weight download logic |
-| `inference.py` | Helper functions for loading model & running prediction |
-| `model.py` | Model architecture / wrapper code |
-| `style.css` | Custom UI styling |
-| `requirements.txt` | Python dependencies |
-| `.gitignore` | Excludes venv, caches, large artifacts |
-
----
-
-## 📦 Dependencies (from `requirements.txt`)
-Key packages (versions resolved by the Space at build time):
-```
-gradio
-torch
-torchvision
-timm
-segmentation-models-pytorch
-numpy
-Pillow
-huggingface-hub
-```
-
-If a future build fails due to incompatible Torch / CUDA combos, pin Torch + torchvision versions explicitly.
-
----
-
-## 🖥 Local Development
-
-```bash
-git clone https://huggingface.co/spaces/Suzyloubna/Bridge.Semantic.Segmentation
-cd Bridge.Semantic.Segmentation
-python -m venv venv
-venv\Scripts\activate  # (Windows)  OR  source venv/bin/activate (Linux/Mac)
-pip install -r requirements.txt
-
-# (Optional) Manually fetch weights if you want to avoid first-run download delay:
-python -c "from huggingface_hub import hf_hub_download; hf_hub_download(repo_id='g-loubna/bridge-unetpp', filename='MILESTONE_090_ACHIEVED_iou_0.9077.pth')"
-
-python app.py
-```
-
-Then open the local Gradio link it prints.
-
----
-
-## 🎨 Visualization Modes
-
-| Mode | Description | Notes |
-|------|-------------|-------|
-| Colored Mask | Pure color rendering per class | Background alpha adjustable |
-| Overlay | Blends original image + color mask | Alpha slider controls blend |
-| Raw Class Indices | Grayscale index visualization | Pixel values scaled 0–255 |
-
----
-
-## 📊 Legend & Statistics
-
-The legend shows:
-- Class ID & name
-- Pixel count
-- Percentage of total area
-
-This is computed after each run using a NumPy bincount on the predicted mask.
-
----
+## Files
+- `app.py` – Gradio interface + weight download
+- `inference.py` – Prediction utilities
+- `model.py` – Model definition/wrapper
+- `style.css` – Custom UI styling
+- `requirements.txt` – Dependencies
 
-## 🧪 Inference Flow
-
-1. Load / download weights (only if not present).
-2. Preprocess image (convert to tensor as defined in `inference.py`).
-3. Forward pass through UNet++ model.
-4. Argmax over class dimension → class index mask.
-5. Post-process & display in selected mode.
-6. Generate stats + optional downloadable raw mask.
-
----
-
-## 🔄 Updating the Weights
-
-If you generate improved weights:
-1. Upload new `.pth` to the model repo (`g-loubna/bridge-unetpp`).
-2. Either keep the same filename OR update:
-   - `WEIGHTS_FILENAME` in `app.py`
-3. Commit & push the Space:
-   ```bash
-   git add app.py
-   git commit -m "Update to new weights file"
-   git push
-   ```
-4. First Space run after that re-downloads the new file.
-
----
-
-## 🧹 Not Tracked (Good)
-
-- `venv/`
-- `__pycache__/`
-- Large weight file (`.pth`) — intentionally excluded; downloaded at runtime.
-
----
-
-## 🛠 Troubleshooting
-
-| Issue | Cause | Fix |
-|-------|-------|-----|
-| Build error: `python:3.1 not found` | Typo in `python_version` | Use `3.10` in front matter |
-| `ModuleNotFoundError: huggingface_hub` | Missing dependency | Add to `requirements.txt`, push |
-| Long startup time | First-time weight download | Wait; subsequent runs are faster |
-| Authentication error when pushing | Wrong token / no write scope | Create token with read+write scopes |
-| Blank UI / model error | Weight not downloaded | Check logs for download failure |
-
----
-
-## 🔐 Tokens & Auth (For Development Pushes)
-
-Generate token: https://huggingface.co/settings/tokens  
-Use as password for `git push` (username = your HF username).
-
----
-
-## 📄 License
-
-Licensed under **Apache-2.0** (see `license` field).  
-Change to `mit` in header and add a LICENSE file if you prefer MIT—ask if you want an MIT license file generated.
-
----
-
-## 📌 Roadmap / Ideas
-
-- Add more structural material classes
-- On-the-fly tiling for huge images
-- Batch processing / ZIP export
-- Per-class IoU benchmark viewer
-- Lightweight quantized model for faster inference
-
----
-
-## 🙌 Acknowledgements
-
-- UNet++ architecture inspiration from academic literature.
-- Hugging Face for model & Space hosting.
-- Gradio for rapid UI prototyping.
-
----
-
-## ✅ Status
-
-If you see this README rendered correctly, the Space config front‑matter is valid.  
-Check the Logs panel if something fails during runtime.
-
----
+## Troubleshooting
+If build fails with a python:3.1 image again, ensure this header appears exactly as shown (no `python_version` line). Push a new commit.
 
-Feel free to open an issue / request enhancements (or ask here) if you need more features.
+## License
+Apache-2.0 (see header). Change `license:` field and add a LICENSE file if you prefer another license.