Structured VQA
Participants must answer True/False Questions (TFQ) and Multiple-Choice Questions (MCQ) with binary or labeled outputs.
Multimedia/CORE Unranked
TRIDENT 2026
[ACMMM GC] TRIDENT: Tri-modal Deepfake Perception, Detection, and Hallucination Grand Challenge
TRIDENT@ACM MM 2026 is a Grand Challenge focused on tri-modal deepfake perception, detection, and hallucination analysis, addressing the limitations of black-box detection systems by requiring models to provide interpretable, grounded reasoning across image, video, and audio modalities. It aims to advance forensic AI toward accountability and transparency by evaluating not just accuracy but also the reliability of model explanations.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Type-A Open-Ended Questions
Given a known manipulated sample, participants must provide a structured description of observable artifacts and manipulation evidence.
Type-B Open-Ended Questions
Given an unknown sample, participants must output an authenticity label ('Likely Authentic' or 'Likely Manipulated') and a short reasoning paragraph.
Conference Paper
Winning teams are invited to submit a paper describing their method to the ACM MM 2026 main conference, subject to review and on-site presentation.
Research areas in scope
01
Tracks
ImageVideoAudio
Policies worth checking twice
- Each team must have a responsible team leader with formal affiliation (faculty, researcher, or staff); students may not serve as team leaders unless approved.
- Each participant may only be a member of one team per track.
- Teams must register via the official form and receive approval before accessing the dataset.
- Maximum of 3 submissions per day per track; excess submissions are rejected.
- Submissions must be in JSON format with specific naming convention: {teamname}_{track}.json.
- Each submission must be accompanied by a 1–2 page abstract in PDF format ({teamname}_abstract.pdf) sent via email.
- Submissions without a valid abstract will appear on the leaderboard but are ineligible for awards.
- Use of external datasets, pretrained models, LLMs, and MLLMs is permitted but must be fully disclosed in the abstract.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026