Awesome

MvDeCor

This is an official code release of

MvDeCor: Multi-view Dense Correspondence Learning for Fine-grained 3D Segmentation

Gopal Sharma, Kangxue Yin, Subhransu Maji, Evangelos Kalogerakis, Or Litany and Sanja Fidler <img src=docs/teaser.png width="1024">

Requirements

• Python 3.9 is supported.
• Pytorch 1.5.1.
• This code is tested with CUDA 10.1 toolkit
• Use the following script to install conda environment

bash install.sh

Dataset download and processing

Use the following script to download and process the dataset

bash dataset_process.sh

update categories to categories you want.

Training and testing

For pretraining, run the following script (requires 2 GPUS):

bash partnet.sh

For training few shot segmentation on partnet dataset, run the following script (require 1 GPU):

bash partnet_seg.sh

Note that test is automatically done in the code, after training is completed.

Citation

@inproceedings{mvdecor2022,
    title={MvDeCor: Multi-view Dense Correspondence Learning for Fine-grained 3D Segmentation},
    author={Gopal Sharma and  Kangxue Yin and  Subhransu Maji and  Evangelos Kalogerakis and  Or Litany and Sanja Fidler},
    booktitle={Proceedings of the European Conference on Computer Vision Workshops (ECCV)},
    year={2022}
}