LibreRFDETRm-cls
LibreRFDETRM classification model trained on imagenette (10-class ImageNet subset).
Model Details
| Property | Value |
|---|---|
| Architecture | LibreRFDETRM |
| Task | Image Classification |
| Input size | 224×224 |
| Classes | 10 |
| Top-1 (imagenette val) | 97.45% |
| Top-5 (imagenette val) | — |
| License | APACHE-2.0 |
Classes
| ID | Name |
|---|---|
| 0 | tench |
| 1 | English springer |
| 2 | cassette player |
| 3 | chain saw |
| 4 | church |
| 5 | French horn |
| 6 | garbage truck |
| 7 | gas pump |
| 8 | golf ball |
| 9 | parachute |
Usage
from libreyolo import LibreRFDETR
model = LibreRFDETR("LibreRFDETRm-cls.pt", task="classify")
result = model.predict("image.jpg")
print(result) # top-1 class and confidence
Training
- Dataset: imagenette160 (train: 9,469 images · val: 3,925 images · 10 classes)
- Epochs: 30
- Optimizer: AdamW
- Scheduler: Warm cosine (5 warmup epochs)
- Augmentation: RandomResizedCrop + TrivialAugmentWide + RandomErasing
- Transfer: pretrained backbone (COCO detection weights)
Limitations
These weights are trained on imagenette (10 classes, ~9.5k images), a fast.ai benchmark subset of ImageNet. They serve as a functional demo of the classification pipeline. For production use we recommend fine-tuning on your own dataset. A full ImageNet-1k training run is planned when compute budget allows.
License
APACHE-2.0 — see LICENSE. The imagenette dataset is derived from ImageNet; the original ImageNet terms apply to the training data.