Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.md

Refine background cues

Download the following pre-trained models GoogleDrive | BaiduYun (pwd: 27p5) into dataset/pretrained folder.

  1. Adopt the extracted background cues as supervision signal to train PoolNet:
OMP_NUM_THREADS=16 CUDA_VISIBLE_DEVICES=0 python3 main_voc.py --arch resnet --mode train --train_root /path/to/your/dataset/VOC2012/ --pseudo_root ../experiments/predictions/path/to/your/background/cues/

the models and visual results will be saved as follow:

├── results/
|   ├── run-xx
|   |   ├—— models
|   |   |   ├—— xxx.pth
├── visual_results
  1. Generate the refined background cues using PoolNet
OMP_NUM_THREADS=16 CUDA_VISIBLE_DEVICES=0 python3 main_voc.py --arch resnet --mode test --train_root /path/to/your/dataset/VOC2012/ --model ./results/path/to/your/saved/model --sal_folder ./results/refined_background_cues

The refined background cues will be saved as follow:

├── results/
|   ├── refined_background_cues
|   |   ├—— xxxx.png

Please Note that we just follow PoolNet train a saliency detector so the generated results actually activate the foreground (white). (1 - results/255) to obtain the real background cues.

Reference

If you are using our code, please consider citing our paper.

@InProceedings{Xie_2022_CVPR,
    author    = {Xie, Jinheng and Xiang, Jianfeng and Chen, Junliang and Hou, Xianxu and Zhao, Xiaodong and Shen, Linlin},
    title     = {{C2AM}: Contrastive Learning of Class-Agnostic Activation Map for Weakly Supervised Object Localization and Semantic Segmentation},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2022},
    pages     = {989-998}
}
@article{xie2022contrastive,
  title={Contrastive learning of Class-agnostic Activation Map for Weakly Supervised Object Localization and Semantic Segmentation},
  author={Xie, Jinheng and Xiang, Jianfeng and Chen, Junliang and Hou, Xianxu and Zhao, Xiaodong and Shen, Linlin},
  journal={arXiv preprint arXiv:2203.13505},
  year={2022}
}

This repository was modified from PoolNet.