Search papers, labs, and topics across Lattice.
The MAGMaR 2026 Shared Task evaluated advancements in multimodal augmented generation through video retrieval and grounded article generation, attracting participation from multiple teams. Notably, all 17 systems submitted for the video retrieval task surpassed a baseline established by the previous year's winner, demonstrating significant progress in this area. Additionally, the generation task yielded 16 systems from 4 teams, with each team producing at least one report that received top human ratings, underscoring the competitive nature and effectiveness of the submissions.
Every team in the generation task produced at least one report deemed the best by human annotators, highlighting a leap in multimodal generation quality.
This overview paper presents the results of the shared task for the second workshop on Multimodal Augmented Generation via Multimodal Retrieval (MAGMaR). In this shared task participants submitted systems focused on either (i) video retrieval or (ii) grounded generation of articles given retrieved videos. Teams could submit to either task. For the retrieval task, we had 2 participating teams that submitted a total of 17 systems -- all of which beat a baseline derived from the winner of last year's shared task. On the generation side, we had 4 teams submit 16 systems. All teams had at least one generated report that was labeled the best by a human annotator.
OpenAI