| title: Reachy Mini Object Detector | |
| emoji: 👁️ | |
| colorFrom: red | |
| colorTo: blue | |
| sdk: static | |
| pinned: false | |
| short_description: Real-time object detection with head tracking | |
| tags: | |
| - reachy_mini | |
| - reachy_mini_python_app | |
| # Reachy Mini Object Detector | |
| Real-time object detection app for [Reachy Mini](https://github.com/pollen-robotics/reachy_mini) using TensorFlow Hub's SSD MobileNet V2 model. | |
| ## Features | |
| - **Live object detection** via the robot's camera using SSD MobileNet V2 (~10 FPS on CPU) | |
| - **Head tracking** — the robot follows the largest detected object with smooth interpolation | |
| - **Antenna wiggle** — antennas wiggle when a new object class is first spotted | |
| - **Web UI** — live annotated video feed with detection list, FPS counter, and tracking toggle | |
| ## How it works | |
| 1. The camera captures frames and feeds them to the TF Hub object detector | |
| 2. Detected objects are drawn as bounding boxes on the video stream | |
| 3. The robot's head smoothly tracks the largest object in frame | |
| 4. A web dashboard at `http://0.0.0.0:8042` shows the live feed and controls | |
| ## Installation | |
| Install from the Reachy Mini dashboard or manually: | |
| ```bash | |
| pip install git+https://huggingface.co/spaces/backtoengineering/reachy_mini_object_detector | |
| ``` | |
| ## Dependencies | |
| - `reachy-mini` | |
| - `tensorflow >= 2.15` | |
| - `tensorflow-hub >= 0.16` | |
| - `opencv-python >= 4.8` | |
| - `numpy` |