Spaces:
Running on Zero
Running on Zero
| title: CrowdCounter | |
| emoji: 👥 | |
| colorFrom: purple | |
| colorTo: indigo | |
| sdk: gradio | |
| sdk_version: 6.19.0 | |
| python_version: '3.12' | |
| app_file: app.py | |
| pinned: false | |
| short_description: Count people in aerial/crowd photos with a dot overlay | |
| # Crowd Counter | |
| Upload an aerial or dense crowd photo; get back the image with a dot on every detected | |
| head and the total count printed underneath. | |
| - **Model:** fine-tuned PET (Point-quEry Transformer), point-localization crowd counter. | |
| - **Inference:** native-resolution tiling for dense-scene recall; adjustable head-confidence | |
| threshold to recover packed-core heads. | |
| - **Hardware:** ZeroGPU (GPU attached per request via `@spaces.GPU`). | |
| Accuracy is best on higher-resolution stills; expect ±10–15% on very dense scenes. The dot | |
| overlay is the audit trail — check whether dots track heads. | |