The first Learning and Mining with Noisy Labels Challenge is a part of IJCAI-ECAI 2022. The dataset is CIFAR-N, a recently collected human-annotations for CIFAR. The goal of the challenge is to prepare the community shift from synthetic label noise benchmarks to tasks with real human-annotated data. Unlike the webvision challenge, CIFAR-N is small in size, containing a set of easily accessible, easy-to-use, and verifiable datasets, making training and evaluating CIFAR-N very resource-friendly.
This year, we organize two tasks to evaluate the learned knowledge and representation on CIFAR-N:
(1) Image Classification Task.
(2) Label Noise Detection Task.
The participants can form as a group taking part in one or two tasks. Please make sure that you have thoroughly read the task description before joining the task challenge. Each group must submit a report describing their method and discussing the results. The formatting guidelines for the report follow the main track in IJCAI-ECAI 2022. Note that the report should be 2-8 pages which describes your method and technical details. We also require each group to attach the Github link to their code in the report. The code should have a clear description for model selection for each task which is crucial because we will examine the correctness of each method. The registration and report is submitted via this link. The timelines are scheduled as follows.
Congratulations to all winners & runner-ups!!!
ProMix: Combating Label Noise via Maximizing Clean Sample Utility
Authors: Haobo Wang, Ruixuan Xiao, Yiwen Dong, Lei Feng, Junbo Zhao
Affiliation: Zhejiang University & Chongqing University
Focus on Center Crop: Noisy labels learning with Cross attention Image Cropping
Authors: Weichen Yu, Hongyuan Yu, Yan Huang, Dong An, Keji He, Zhipeng Zhang,
Xiuchuan Li, Liang Wang
Affiliation: Institute of Automation, Chinese Academy of Sciences
Noise-Robust Bidirectional Learning with Dynamic Sample Reweighting
Authors: ChenChen Zong, ZhengTao Cao, HongTao Guo, Yun Du, MingKun Xie,
ShaoYuan Li , ShengJun Huang
Affiliation: Nanjing University of Aeronautics and Astronautics,
MIIT Key Laboratory of Pattern Analysis and Machine Intelligence
If you have further questions about the challenge, please contact us via lmnl.challenge@gmail.com.