RF-DETR Small piston parts detector
Detector for 11 pneumatic-piston part/tool classes, trained with RF-DETR 1.3.0 and 512-pixel input.
from rfdetr import RFDETRSmall
model = RFDETRSmall(pretrain_weights="checkpoint_best_total.pth", num_classes=11)
detections = model.predict("image.jpg", threshold=0.5)
Use RGB channel order for NumPy images. labels.json preserves category IDs 1β11; the checkpoint includes 12 class-head slots.
Evaluation
| Split | Precision | Recall | mAP@0.5 |
|---|---|---|---|
| Validation | 0.967438 | 0.870000 | 0.948721 |
| Test | 0.948808 | 0.920000 | 0.957689 |
Metrics come from the training reports; the manuscript reports the validation results. Configuration, environment details, reports and file hashes are included.
Dataset: Piston parts detection. Paper and code: Vision assembly assistance. License: Apache 2.0.