Skip to content

Learning visual representations under the existence of label noise

License

Notifications You must be signed in to change notification settings

ggalan87/noise-robust-learning

Repository files navigation

Noise-robust learning

A framework for learning under the presence of label noise.

Dependencies

The repository is mostly built around pytorch lightning framework and pytorch-metric-learning library.

Parts of the repository includes code from other sources. Every such part has link to its source, e.g. url to the relevant implementation.

The requirements file includes the dependencies of the project, although some may be deprecated.

How to run

The scripts which correspond to the experiments reported in the papers can be found in experiments scripts folder. All experiments are mostly built around lighning package which contains models, datasets etc. Data denoising methods can be found in torch_metric_learning package which is essentially an extension to PML.

TODO

  • Refactor files and packages
  • Provide / improve documentation
  • Remove custom paths from test/example files
  • Migrate to newer dependencies versions
  • Migrate to poetry for dependency management

Relevant publications

This repository contains the code of the following papers.

@inproceedings{galanakis2024noise,
  title={Noise-robust person re-identification through nearest-neighbor sample filtering},
  author={Galanakis, George and Zabulis, Xenophon and Argyros, Antonis A},
  booktitle={2024 IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS)},
  pages={1--8},
  year={2024},
  organization={IEEE}
}

@inproceedings{Galanakis2024,
  author = {Galanakis, George and Zabulis, Xenophon and Argyros, Antonis A},
  title = {Nearest neighbor-based data denoising for deep metric learning},
  booktitle = {International Conference on Computer Vision Theory and Applications (VISAPP 2024)},
  publisher = {Scitepress},
  year = {2024},
  month = {February},
  pages = {595--603}
}  

About

Learning visual representations under the existence of label noise

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages