Inference¶
Run predictions on videos and label files, then dig into the focused topics below.
| Page | Description |
|---|---|
| Running Inference | Run predictions from the CLI |
| Post-Processing Filters | Node-count, confidence, and overlap-NMS filters |
| Python API | Embed inference in Python (Predictor / predict / Outputs) |
| Centroid-Only | Run a centroid model standalone |
| Top-Down Segmentation | Per-instance masks via centroid + crop-mask |
| SAM-Prompted Segmentation | Per-instance masks from poses via Segment Anything |
| Performance | Tune throughput (FP16, torch.compile, workers) |