GUANGZHOU, China, July 04, 2026 (GLOBE NEWSWIRE) — The organizing committee recently announced the official launch of the MARS2 Multimodal Reasoning Competition. Sponsored by Tec-Do and MiniMax, a flagship event held alongside the MARS2 Workshop at the upcoming European Conference on Computer Vision (ECCV 2026). Backed by a total prize pool of $100,000 USD, the challenge aims to bridge vanguard academic research with high-impact, real-world industry applications.
Unlike traditional AI benchmarks confined to closed tasks, the MARS2 competition shifts the paradigm toward complex, open-world reasoning, deep “slow-thinking” capabilities, and long-chain multi-step inferences. The contest centers on three distinct tracks designed to thoroughly test multimodal models in complex, enterprise-level environments:
- MAC (Multimodal Advertisement Comprehension): Evaluates a model’s holistic grasp of video narratives, cross-modal integration, and its capacity for long-context semantic summarization.
- VTG (Video Temporal Grounding): Measures fine-grained temporal alignment and semantic localization within dynamic video streams.
- MDC (Marketing Strategy Decoding and Conversion Analysis): Challenges models to decode psychological triggers, brand placements, and user retention mechanisms within commercial contexts.
Bridging Vanguard Research with Real-World Industry Value
For academic researchers, MARS2 offers a robust, challenging benchmark to pioneer open-world generalization and testing-time reasoning. For industry practitioners, it serves as a critical sandbox to transition raw computing power into productive commercial results—specifically within digital marketing, intelligent decision-making, and structural cost-optimization.
The workshop features an elite lineup of invited keynote speakers, including:
Paul Liang (Massachusetts Institute of Technology)
Mihaela van der Schaar (University of Cambridge)
Louis-Philippe Morency (Carnegie Mellon University)
Call for Papers (CFP)
The MARS2 Workshop at ECCV 2026 warmly invites novel submissions spanning academia and industry. Topics of interest include, but are not limited to: Multimodal Reasoning, Long-Chain Inference, Zero-Shot Generalization, Agentic Systems, Neural-Symbolic Models, and Robustness Benchmarking.
Key Deadlines & Timeline:
Submission Deadline: August 1, 2026
Acceptance Notification: August 8, 2026
Camera-Ready Deadline: August 12, 2026
Submission Portal: OpenReview Portal
Distinguished Organizing Committee
The event is jointly stewarded by leading minds across global research hubs and tech sectors:
Depin Chen (Tec-Do Technology)
Kyoung Mu Lee (Seoul National University)
Chen Change Loy (Nanyang Technological University)
David Clifton (University of Oxford)
Xinwei Long (Tsinghua University)
Luc Van Gool (INSAIT Sofia University)
Event Logistics & How to Join
Event Date: September 8–9, 2026
Location: Malmö, Sweden
Challenge Platform: Registration, submissions, and live leaderboards will be hosted on the EvalAI Challenge Page.
To explore the frontiers of multimodal intelligence alongside global peers at ECCV 2026, visit the official channels below to register or submit your work.
Official Website: https://mars2workshop.github.io/eccv2026/
Inquiries & Contact: [email protected]
Company: Tec-Do Technology Co., Limited
Contact Person: Monica Guan
Email: [email protected]
Website: https://www.tec-do.com/
Telephone: +8613924135206
City: Guangzhou, China
A photo accompanying this announcement is available at https://www.globenewswire.com/NewsRoom/AttachmentNg/6dd239ca-745a-4a5c-a985-99b3a4e2f109
