Live Grid
Run 4 default pose models in a stable round-robin queue. MoveNet is available but off by default.
Solo Test
Run only the selected model to measure peak performance and tracking quality.
Fair Benchmark
Capture 30 frames and process each hosted model individually for a fair comparison.
Start camera to begin benchmark
| Model | Avg | P95 | FPS | Hit Rate | Avg People |
|---|
Silhouette
Evaluate silhouette quality and processing speed independently from pose estimation.
Licenses
Licenses for libraries and models used in this tool.
| Library / Model | License | Author |
|---|---|---|
| @mediapipe/tasks-vision | Apache-2.0 | |
| MediaPipe Pose Landmarker Lite | Apache-2.0 | |
| MediaPipe Selfie Segmenter | Apache-2.0 | |
| onnxruntime-web | MIT | Microsoft |
| YOLOv8n-Pose | AGPL-3.0 | Ultralytics |
| YOLOX-nano detector | Apache-2.0 | Megvii / OpenMMLab ONNX export |
| MoveNet MultiPose Lightning | Apache-2.0 | |
| RTMPose-s | Apache-2.0 | OpenMMLab |
| RTMO-s | Apache-2.0 | OpenMMLab |
| MODNet | Apache-2.0 | ZHKKKe |
| PP-HumanSeg | Apache-2.0 | PaddlePaddle |
| esbuild | MIT | Evan Wallace |