OrthogonalDet

Code for CVPR 2024 paper Exploring Orthogonality in Open World Object Detection.

<p align="center"> <img src="assets/intro.png" alt="OrthogonalDet" width=60%> </p>

Requirements

• Linux or macOS with Python ≥ 3.8.
• Install PyTorch ≥ 1.9.0, torchvision, Detectron2, timm, and einops.
• Prepare datasets:
  • Download COCO and PASCAL VOC.
  • Convert annotation format using coco_to_voc.py.
  • Move all images to datasets/JPEGImages and annotations to datasets/Annotations.

Getting Started

• Training for open world object detection:

  bash run_owod.sh
  

  Evaluation for open world object detection:

  bash test_owod.sh
  

• Experiment for incremental object detection:

  bash run_iod.sh
  

• Visualize the results:

  python demo.py -i LIST_OF_IMAGES
  

• Note that we are using an ImageNet pre-trained backbone. To switch to a DINO pre-trained backbone, please download the model weights and then follow these instructions.

Results

The following results were obtained with four NVIDIA 2080 Ti GPUs, using the checkpoints at this link.

• Open world object detection on M-OWODB and S-OWODB:

  

• Incremental object detection on PASCAL VOC:

  

Citation

If you find this code useful, please consider citing:

@inproceedings{sun2024exploring,
  title={Exploring Orthogonality in Open World Object Detection},
  author={Sun, Zhicheng and Li, Jinghan and Mu, Yadong},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={17302--17312},
  year={2024}
}

Acknowledgement

Our implementation is based on RandBox which uses Detectron2 and Sparse R-CNN.