Welcome!
Welcome to "Erasing the Invisible" at NeurIPS 2024—a cutting-edge competition designed to rigorously stress-test image watermarks. This competition is divided into two tracks: Black Box Track and Beige Box Track. It aims to validate the robustness of image watermarks under varying visibility conditions and attacker knowledge. Competitors will attempt to remove invisible watermarks while maintaining the quality of the images. Evaluations will be based on two criteria: the effectiveness of watermark removal and the preservation of image quality. This competition is inspired by the WAVES benchmark.
Prizes: There is a $7000 prize pool!
Join our Slack now!
News
- Nov 5: Due to the Codabench issue, we extend the submission deadline to Nov 7.
- Oct 22: We increase the prize pool to $7,000!
- Sep 24: We release an interactive Demo for this competition on Hugging Face. Try it out!
- Sep 16: The submission phase has started! See Tracks for details.
- Sep 8: We double the prize pool to $6,000!
- Sep 8: The submission phase has been postponed to Sep 16 - Nov 5.
- Aug 27: The start of the submission phase has been postponed to Sep 9.
- July 31: The warm-up phase has started. See Getting Started for details.
- July 31: Registration has opened.
Registration
Please register for this competition by filling out this Google Form. We will email you with every update, ensuring you won't miss anything!Important Dates
- July 31: Registration opens
- July 31 - Sep 15: Warm-up phase
- Sep 16 -
Nov 5Nov 7:: Submission phase Nov 5Nov 7: Registration closes, submission closes- Dec 15: Workshop day at NeurIPS'24 (Winners will be announced in the Workshop)
Rules
- Registration: This competition is open to all. Participants must be registered with a valid email address before submitting an entry. One account per participant. Double registration is not allowed. Every participant can only join one team. Teams may participate in multiple tracks. There is no limit on team size.
- Submission Requirements: 1) Participants receive a set of watermarked images. They then attack these images and upload them. 2) The submitted images must follow the guidelines of naming and formatting. Please find details on the submission platform. 3) The exploitation of loopholes (such as the leakage of underlying model) is prohibited. Participants are encouraged to report any detected loopholes. 4) During the competition period, we encourage general discussions, but teams are not allowed to share their code or submissions with other teams.
- Prize Eligibility: We will provide monetary prizes to the top teams for each track, subject to scrutinized legitimacy of their methods. For prize eligibility, leading teams must at least provide their methods and code to the organizers for legitimacy checking. Winning teams will be selected by human judges to avoid any hacks on quality metrics. Winning teams will be invited to give a talk at the competition workshop. University System of Maryland, including University of Maryland College Park Employees are not eligible for the prize.
- Rule Adaptation Clause: By registering, participants agree that adjustments may be made in response to unexpected situations. We will strive for fair resolutions to address any unforeseen issues.
Contact
- To contact the organizers, please email erasinginvisible@googlegroups.com
- Please join our Slack to receive the latest announcements, engage in discussions, find teammates, and ask questions.
Sponsors
Organizers
(* indicates Lead Organizers)
UMD
Yale
UMD
UMD
UMD
UMD
UMD
UMD
UMD
UMD
UC Berkeley
JHU
Yale
CMU
UCSD
JHU
UMD
UMD
UMD
@inproceedings{
an2024waves,
title={{WAVES}: Benchmarking the Robustness of Image Watermarks},
author={Bang An and Mucong Ding and Tahseen Rabbani and Aakriti Agrawal and Yuancheng Xu and Chenghao Deng and Sicheng Zhu and Abdirisak Mohamed and Yuxin Wen and Tom Goldstein and Furong Huang},
booktitle={Forty-first International Conference on Machine Learning},
year={2024},
url={https://openreview.net/forum?id=URtUYfC3GA}
}