On an RTX 5070 Ti in PyTorch, PicoDet averages 5.1 mAP points higher than YOLOX at matched parameter counts, winning both nearest-params pairings in the sub-5M tier. YOLOX is 21.3% faster on average and is the only family that scales past 3M parameters: its largest variant reaches 56.3 mAP against PicoDet's top of 44.1. Pick PicoDet for the smallest models, YOLOX when you need more accuracy or throughput.
PicoDet fields 3 measured variants and YOLOX fields 6, all on the same COCO val2017 protocol on an RTX 5070 Ti. PicoDet tops out at 3.31M parameters, so only its tiny end overlaps YOLOX, giving 2 matched pairs.
| Model | Family | Params (M) | mAP@50-95 | FPS |
|---|---|---|---|---|
| YOLOX-Nano | YOLOX | 0.9 | 2876.0 | 53.6 |
| PicoDet-S | PicoDet | 1.0 | 3037.0 | 50.0 |
| PicoDet-M | PicoDet | 2.1 | 3789.0 | 45.1 |
| PicoDet-L | PicoDet | 3.3 | 4413.0 | 39.5 |
| YOLOX-Tiny | YOLOX | 5.1 | 3547.0 | 61.6 |
| YOLOX-S | YOLOX | 9.0 | 4427.0 | 50.0 |
| YOLOX-M | YOLOX | 25.3 | 5169.0 | 48.2 |
| YOLOX-L | YOLOX | 54.2 | 5542.0 | 42.8 |
| YOLOX-X | YOLOX | 99.1 | 5627.0 | 39.6 |
Accuracy at matched compute
mAP is shown in percent. Each PicoDet variant is paired with the nearest YOLOX variant by parameter count. Across the 2 overlapping pairings PicoDet averages 5.1 mAP points higher and wins both. PicoDet-S (0.99M) reaches 30.4 mAP against YOLOX-Nano (0.91M) at 28.8. Above 5M parameters PicoDet has no entry: YOLOX-X reaches 56.3 mAP at 99.07M, while PicoDet-L, the largest measured, sits at 44.1.
Speed
Averaged across the 2 matched pairs, YOLOX is 21.3% faster than PicoDet in PyTorch. YOLOX keeps that throughput as it scales, which PicoDet cannot follow past its 3 small variants.
- PicoDet license
- Apache-2.0
- YOLOX license
- Apache-2.0
- Both families
- permissive, cleared for commercial use
Which family to pick
Pick PicoDet when the model must be tiny: it wins both sub-5M pairings and reaches 44.1 mAP at 3.31M parameters. Pick YOLOX when you need higher absolute accuracy or a larger model, up to 56.3 mAP, or the extra throughput. Both ship under Apache-2.0, so licensing does not force the call.
Every number on this page comes from the verified dataset: same 500-image COCO val2017 slice, conf 0.001, IoU 0.6, max 300 detections, pycocotools mAP, identical protocol across all hardware and runtimes. The full protocol is on the methodology page. To rerun this comparison with your own filters, open compare. Accuracy is measured on LibreYOLO retrained checkpoints; other weight sources can yield different values.