Skip to content
Kitbay

huggingface-vision-trainer

Apache-2.0Updated on Nov 11, 202611,334GitHub starsIncludes scripts

Hugging Face's skill for training computer-vision models, mostly on Hugging Face Jobs cloud GPUs.

What the skill teaches

Hugging Face's skill for training computer-vision models, mostly on Hugging Face Jobs cloud GPUs. It covers fine-tuning object detectors such as D-FINE, RT-DETR v2, DETR and YOLOS, image classifiers from timm or Transformers, and SAM or SAM2 for prompted segmentation. The agent prepares COCO-format or mask datasets, sets up Albumentations augmentation, runs training with bundled scripts, measures results with mAP and mAR, and saves the model to the Hub so it outlives the job. Local training is possible too. Jobs need a Hugging Face account with billing and a write token.

Capabilities

  • Fine-tune object detection models on custom datasets
  • Train image classifiers with timm or Transformers
  • Fine-tune SAM or SAM2 for segmentation
  • Evaluate with mAP and mAR and save models to the Hub