Official Pytorch implementation for Addressing out-of-distribution label noise in webly-labelled data.
Paper link: thecvf
The path to the datasets should be set in the mypath.py file
bash train.sh
You can resume a checkpoint using
--resume path/to/checkpoint.pth.tar
The models trained on Webvision can be downloaded here (--seeds 1, 2 ,3): gdrive
The model trained on Webvision can be tested on the Imagenet validation set using
python eval_imagenet.py model1_state_dict.pth.tar
and optionally for an ensemble
python eval_imagenet.py model1_state_dict.pth.tar model2_state_dict.pth.tar
@inproceedings{2022_WACV_DSOS,
title="{Addressing out-of-distribution label noise in webly-labelled data}",
author="Albert, Paul and Ortego, Diego and Arazo, Eric and O{'}Connor, Noel and McGuinness, Kevin",
booktitle="{Winter Conference on Applications of Computer Vision (WACV)}",
year="2022"}