| title: Qwen Wall Segmentation | |
| emoji: 🪄 | |
| colorFrom: purple | |
| colorTo: pink | |
| sdk: gradio | |
| sdk_version: 5.49.1 | |
| app_file: app.py | |
| pinned: false | |
| license: apache-2.0 | |
| short_description: Qwen Wall Segmentation | |
| # Qwen Wall Segmentation | |
| Produces pixel-accurate wall segmentation masks from a single room photo — no segmentation model, no training data, no manual annotation. | |
| The insight is to turn a hard segmentation problem into a trivial one. Generic wall segmentation is difficult: walls have no consistent shape, blend into ceilings and floors, and vary wildly in color and texture. Rather than fight that, this tool uses [Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511) to repaint only the walls a flat, perfectly uniform color — every wall forced to the same hue, saturation, and brightness — while leaving furniture, floors, windows, shadows, and lighting untouched. The model handles the semantic understanding of "what is a wall"; the uniform recolor then makes those pixels cleanly separable by color. | |
| This sidesteps the usual failure modes of color-based segmentation (lighting gradients, shadows, multi-colored walls), because the recolor normalizes all of them away before extraction. | |
| ## Outputs | |
| For each uploaded image you get three results: the recolored edit, the binary wall mask, and an overlay of the mask on your original image. | |
| ## Use cases | |
| - Bootstrapping wall-segmentation training sets | |
| - Interior virtual-repainting and color visualization | |
| - AR room staging | |
| - Cheap pseudo-labeling for downstream models | |
| Built for the **Build Small Hackathon** (model ≤ 32B params). |