Skip to content
/ CLRec Public

Pytorch implementation for "The Surprising Positive Knowledge Transfer in Continual 3D Object Shape Reconstruction"

Notifications You must be signed in to change notification settings

rehg-lab/CLRec

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

62 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

The Surprising Positive Knowledge Transfer in Continual 3D Object Shape Reconstruction

In this work we investigate continual learning of reconstruction tasks which surprisingly do not suffer from catastrophic forgetting and exhibit positive forward knowledge transfer. In addition, we provide a novel analysis of knowledge transfer ability in CL. We further show the potential of using the feature representation learned in 3D shape reconstruction to serve as a proxy task for classification. Link to our paper and link to our project webpage.

This repository consists of the code for reproducing CL of 3D shape reconstruction, proxy task and autoencoder results in the main text, and YASS and dynamic representation tracking in the appendix.

Training and evaluating Single Object 3D Shape Reconstruction and proxy task

Follow instructions in CL3D README

Training and evaluating autoencoder

Follow instructions in Autoencoder README

Training and evaluating YASS

Follow instructions in YASS README

Dynamic Representation Tracking

Follow instructions in DyRT README

Citing

@misc{thai2021surprising,
    title={The Surprising Positive Knowledge Transfer in Continual 3D Object Shape Reconstruction},
    author={Anh Thai and Stefan Stojanov and Zixuan Huang and Isaac Rehg and James M. Rehg},
    year={2021},
    eprint={2101.07295},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}