hustvl/YOLOS
A vision transformer model adapted for object detection without task-specific architectural modifications, published at NeurIPS 2021.
Not currently ranked — collecting fresh signals.
star history
YOLOS demonstrates that vanilla Vision Transformers pre-trained on image classification can transfer to object detection by adding detection tokens and using a set-based Hungarian matching loss. The project studies the transferability of ImageNet-pretrained ViTs to the COCO detection benchmark, including experiments with self-supervised MoCo-v3 pre-training. The implementation is integrated into HuggingFace Transformers for easy use.
Frequently asked
- What is hustvl/YOLOS?
- A vision transformer model adapted for object detection without task-specific architectural modifications, published at NeurIPS 2021.
- Is YOLOS open source?
- Yes — hustvl/YOLOS is open source, released under the MIT license.
- What language is YOLOS written in?
- hustvl/YOLOS is primarily written in Jupyter Notebook.
- How popular is YOLOS?
- hustvl/YOLOS has 904 stars on GitHub.
- Where can I find YOLOS?
- hustvl/YOLOS is on GitHub at https://github.com/hustvl/YOLOS.