Official PyTorch implementation for the CVPR 2022 paper: "Amodal Segmentation through Out-of-Task and Out-of-Distribution Generalization with a Bayesian Model"
conda env create -f environment.yml
conda activate amodal
bash download.sh
download.sh
cannot be executed properly, please identify the missing directory and rerun the wget
command for the corresponding zip file. If the issue persists, please refer to the download.sh
description below.Table 1: change the file Code/configs.py
to set TABLE_NUM = 1
and MODEL_TYPE = 'ML'
or MODEL_TYPE = 'E2E'
and run the command below.
Table 2: change the file Code/configs.py
to set TABLE_NUM = 2
and MODEL_TYPE = 'ML'
or MODEL_TYPE = 'E2E'
and run the command below.
Table 3: change the file Code/configs.py
to set TABLE_NUM = 3
and MODEL_TYPE = 'ML'
or MODEL_TYPE = 'E2E'
and run the command below.
cd Code
python3 run_experiment.py
download.sh
DescriptionDownload pretrained model weights from here, unzip Models.zip
and place the folder as /Models/
.
Download RPN results used for evaluatiooon from here, unzip RPN_results.zip
and place the folder as /RPN_results/
.
Download Occluded Vehicle Dataset from here, unzip Occluded_Vehicles.zip
and place the folder as /Dataset/Occluded_Vehicles/
.
Download KINS Dataset from here, unzip kitti.zip
and place the folder as /Dataset/kitti/
.
Download COCOA Dataset from here, unzip COCO.zip
and place the folder as /Dataset/COCO/
. Additionally, download COCO data train2014 and val2014 and place the folders as /Dataset/COCO/train2014/
and /Dataset/COCO/val2014/
.
Please cite the following papers if you use the code directly or indirectly in your research projects.
@article{sun2021amodal,
title=Amodal Segmentation through Out-of-Task and Out-of-Distribution Generalization with a Bayesian Model},
author={Sun, Yihong and Kortylewski, Adam and Yuille, Alan},
journal={arXiv preprint arXiv:2010.13175},
year={2021}
}