xingyizhou/UniDet
A multi-dataset object detector trained on COCO, Objects365, OpenImages, and Mapillary using Cascade-RCNN with a learned unified label space.
Not currently ranked — collecting fresh signals.
star history
This is a computer vision research project that trains a unified object detector across multiple large-scale datasets. The detector uses Cascade-RCNN architecture with backbones like ResNeSt200 to achieve state-of-the-art performance on detection benchmarks. The key innovation is learning a unified label space that enables zero-shot transfer to novel datasets outside the training domain.
Frequently asked
- What is xingyizhou/UniDet?
- A multi-dataset object detector trained on COCO, Objects365, OpenImages, and Mapillary using Cascade-RCNN with a learned unified label space.
- Is UniDet open source?
- Yes — xingyizhou/UniDet is an open-source project tracked on heatdrop.
- What language is UniDet written in?
- xingyizhou/UniDet is primarily written in Python.
- How popular is UniDet?
- xingyizhou/UniDet has 517 stars on GitHub.
- Where can I find UniDet?
- xingyizhou/UniDet is on GitHub at https://github.com/xingyizhou/UniDet.