The PyTorch implementation of DiscoBox: Weakly Supervised Instance Segmentation and Semantic Correspondence from Box Supervision.
@article{lan2021discobox,
title={DiscoBox: Weakly Supervised Instance Segmentation and Semantic Correspondence from Box Supervision},
author={Lan, Shiyi and Yu, Zhiding and Choy, Christopher and Radhakrishnan, Subhashree and Liu, Guilin and Zhu, Yuke and Davis, Larry S and Anandkumar, Anima},
journal={arXiv preprint arXiv:2105.06464},
year={2021}
}[ Paper ]
This repository is the implementation of evaluating DiscoBox on PF-Pascal dataset.
This implementation is based on SCOT
conda create -n scot python=3.6
conda activate scot
cat /usr/local/cuda/version.txt
conda install pytorch=1.4.0 torchvision cudatoolkit=10.0 -c pytorch (if CUDA 10)
conda install pytorch=1.4.0 torchvision cudatoolkit=9.0 -c pytorch (if CUDA 9)
conda install -c anaconda scikit-image
conda install -c anaconda pandas
conda install -c anaconda requests
pip install gluoncv-torch
Download the pretrained weights from this link.
Results on PF-PASCAL with res101:
python evaluate_map_CAM.py --dataset pfpascal --thres img --backbone resnet101 --hyperpixel '(2,22,24,25,27,28,29)' --sim OTGeo --exp1 1.0 --exp2 0.5 --eps 0.05 --gpu 0 --classmap 1 --split test --alpha 0.05
python evaluate_map_CAM.py --dataset pfpascal --thres img --backbone resnet101 --hyperpixel '(2,22,24,25,27,28,29)' --sim OTGeo --exp1 1.0 --exp2 0.5 --eps 0.05 --gpu 0 --classmap 1 --split test --alpha 0.10
python evaluate_map_CAM.py --dataset pfpascal --thres img --backbone resnet101 --hyperpixel '(2,22,24,25,27,28,29)' --sim OTGeo --exp1 1.0 --exp2 0.5 --eps 0.05 --gpu 0 --classmap 1 --split test --alpha 0.15
| Method | Backbone | PCK@0.05 | PCK@0.1 | PCK@0.15 |
|---|---|---|---|---|
| SCOT | ResNet-101 | 63.2 | 85.4 | 92.8 |
| DiscoBox | ResNet-101 | 62.7 | 85.6 | 93.5 |
Please look at this link for more details
