detic-detecting-20k-classes-using-image-level-supervision

Overview

• Original paper: https://arxiv.org/abs/2201.02605
• Official GitHub: https://github.com/facebookresearch/Detic

Detic provides a gain of 2.4 mAP for all classes and 8.3 mAP for novel classes on the open-vocabulary LVIS benchmark

Watch YouTube Explainer Video

Features

• Detects any class given class names (using CLIP).
• We train the detector on ImageNet-21K dataset with 21K classes.
• Cross-dataset generalization to OpenImages and Objects365 without finetuning.
• State-of-the-art results on Open-vocabulary LVIS and Open-vocabulary COCO.
• Works for DETR-style detectors.

Demo

Click here to run a demo

References

@inproceedings{zhou2021detecting,
  title={Detecting Twenty-thousand Classes using Image-level Supervision},
  author={Zhou, Xingyi and Girdhar, Rohit and Joulin, Armand and Kr{\"a}henb{\"u}hl, Philipp and Misra, Ishan},
  booktitle={arXiv preprint arXiv:2201.02605},
  year={2021}
}