Loading…
Fine-Tuning SOTA Object Detection Models on Real-World Datasets
Arina Belova
- Source
- Jetbrains
- Published
- Added to Yomu
Summary
This practical comparison examines how YOLO12, YOLO26, and RF-DETR behave when moved from COCO to specialized real-world object-detection datasets, and how fine-tuning changes the outcome. The authors first evaluate six pretrained checkpoints on COCO val2017, then use cable-damage, bone-fracture, and soda-bottles datasets from RF100-VL for zero-shot testing, fine-tuning, and accuracy/latency comparison. Because the model implementations require incompatible dependencies, the workflow uses one PyCharm project with three isolated uv environments and remote GPU execution; the reported latency stays in native frameworks rather than TensorRT. All models perform well on COCO, but direct performance on specialized data falls close to zero, while 10 fine-tuning epochs adapt the detectors well to cable damage and soda bottles. Bone-fracture remains difficult despite fine-tuning, with RF-DETR performing best, indicating that domain shift may require more data, longer training, or domain-specific pretraining.
Context
Pretrained detectors are generally trained on COCO, which contains everyday scenes and 80 common categories. Specialized deployment data, including industrial cable damage, X-ray bone fractures, and heavily occluded soda bottles, can contain unseen classes and differ substantially in imagery, texture, viewpoint, and object density.
Approach / What changed
The work evaluates six YOLO12, YOLO26, and RF-DETR checkpoints on COCO val2017, then tests and fine-tunes them on three RF100-VL datasets. It uses separate uv environments for each model family within one PyCharm project, remote GPU execution, and native-framework inference before comparing accuracy and latency.
Takeaways
- On COCO val2017, RF-DETR Base recorded the highest mAP50-95 at 0.5325, while YOLO26-N had the lowest listed latency at 12.3 ms.
- Native-framework latency is not directly comparable with optimized paper benchmarks because the run used different hardware and did not compile models with TensorRT.
- Soda-bottle predictions were generally tight and accurate; cable-damage localization was less consistent, and fewer than half of bone-fracture images had an identified break.