Awn Studio Model Zoo
A curated set of segmentation checkpoints for Awn Studio. The standard setup prepares only YOLO11N; the two heavier models are optional downloads.
| Model | Complete-awn recall ↑ | Fragmentation rate ↓ | Structural coverage ↑ | Speed | Download size |
|---|---|---|---|---|---|
| YOLO11N | 82.7% | 28.2% | 91.7% | 2.29 s/page | 5.7 MB |
| RF-DETR Seg XL | 81.6% | 17.6% | 92.1% | 12.84 s/page | 144.5 MB |
| Mask2Former Swin-L | 83.2% | 20.1% | 95.3% | 32.96 s/page | 826.0 MB |
All three structural metrics use the same complete-awn benchmark protocol, and speed is measured with the same full-page timing benchmark.
These three models cover the main trade-offs: YOLO11N for routine use, RF-DETR for lower fragmentation, and Mask2Former for maximum structural coverage.
The repository also retains additional comparison checkpoints from development, but they are not part of the main Awn Studio model library.
Main files
yolo11n-default.ptrfdetr-seg-xl.pthmask2former-swin-l/(model.safetensors,config.json,preprocessor_config.json)
Project repository: https://github.com/Ushiochanii/wheat-awn-phenotyping