Model Performance
| Metric | Final | Best |
|---|---|---|
| mAP@50-95 (B) | 0.8434 | 0.8446 |
| mAP@50 (B) | 0.9687 | 0.9865 |
| Precision (B) | 0.9843 | 0.9949 |
| Recall (B) | 0.9511 | 0.9651 |
Training Curves
| mAP@50-95 | mAP@50 |
 |
 |
| Precision | Recall |
 |
 |
Training Hyperparameters
General
| Parameter | Value |
|---|---|
| Base model | yolo26s.pt |
| Task | detect |
| Image size | 1280 |
| Epochs | 1200 |
| Batch size | 16 |
| Nominal batch size (nbs) | 64 |
| Device | 1,0 (2× GPU) |
| Workers | 8 |
| Pretrained | true |
| AMP | true |
| Cache | true |
| Seed | 0 |
Optimizer
| Parameter | Value |
|---|---|
| Optimizer | MuSGD |
| lr0 | 0.01 |
| lrf | 0.005 |
| Cosine LR | true |
| Momentum | 0.937 |
| Weight decay | 5e-4 |
| Warmup epochs | 15 |
| Warmup momentum | 0.8 |
| Warmup bias lr | 0.1 |
| Patience (early stop) | 10000 |
Loss
| Parameter | Value |
|---|---|
| box | 7.5 |
| cls | 0.5 |
| dfl | 1.5 |
Augmentation
| Parameter | Value |
|---|---|
| Mosaic | 0.7 |
| Close mosaic | 30 |
| fliplr | 0.5 |
| flipud | 0.0 |
| translate | 0.1 |
| scale | 0.7 |
| shear | 5 |
| degrees | 0.0 |
| perspective | 3e-4 |
| hsv_h | 0.015 |
| hsv_s | 0.7 |
| hsv_v | 0.4 |
| copy_paste | 0.0 |
| mixup | 0.0 |
| cutmix | 0.0 |
| erasing | 0 |
Inference / NMS
| Parameter | Value |
|---|---|
| IoU threshold | 0.7 |
| Max detections | 300 |
- Downloads last month
- 116