Negative Prototypes Guided Contrastive Learning for Weakly Supervised Object Detection
If this paper is useful for your research, please kindly cite our paper Negative Prototypes Guided Contrastive Learning for Weakly Supervised Object Detection:
@inproceedings{zhang2023negative,
title={Negative Prototypes Guided Contrastive Learning for Weakly Supervised Object Detection},
author={Zhang, Yu and Zhu, Chuang and Yang, Guoqing and Chen, Siqi},
booktitle={Joint European Conference on Machine Learning and Knowledge Discovery in Databases},
pages={36--51},
year={2023},
organization={Springer}
}
mkdir -p datasets/{coco/voc}
datasets/
├── voc/
│ ├── VOC2007
│ │ ├── Annotations/
│ │ ├── JPEGImages/
│ │ ├── ...
│ ├── VOC2012/
│ │ ├── ...
├── coco/
│ ├── annotations/
│ ├── train2014/
│ ├── val2014/
│ ├── train2017/
│ ├── ...
├── ...Download .pkl file
mkdir proposal
proposal/
├── SS/
│ ├── voc
│ │ ├── SS-voc07_trainval.pkl/
│ │ ├── SS-voc07_test.pkl/
│ │ ├── ...
├── MCG/
│ ├── voc
│ │ ├── ...
│ ├── coco
│ │ ├── MCG-coco_2014_train_boxes.pkl/
│ │ ├── ...
├── ...Please run the follow code to test the model.
python -m torch.distributed.launch --nproc_per_node={NO_GPU} tools/{file}.py
--config-file "configs/{config_file}.yaml"
OUTPUT_DIR {output_dir}
nms {nms threshold}
lmda {lambda value}
iou {iou threshold}
temp {temperature}Some parts of the code are borrowed from wetectron