10+ Years of MADIMA · MADIMA @ ACCV 2026

Multimodal AI for Dietary Intelligence, Management, and Assessment

For more than a decade, MADiMa has brought together researchers advancing AI, multimedia, sensing technologies, food, nutrition, and health. Join us as we shape the future of multimodal food intelligence.

Submit a Paper View Program

Venue

Osaka, Japan

Date

Dec 14–15, 2026 (TBC)

Next Deadline

September 22, 2026

Workshop Overview

Over the past decade, MADIMA has established itself as a recognized forum for multimedia-assisted dietary assessment. Many specialized but fragmented initiatives have since emerged, creating strong momentum and a growing need for shared benchmarks, standardization practices, interoperable frameworks, and interdisciplinary collaboration. MADIMA2026 aims to evolve toward a broader workshop on AI-driven methods for nutrition, health, and human-centered assistance. Building on its foundations in computer vision for dietary assessment, it expands into multimodal foundation models, including vision-language models (VLMs), vision-based retrieval-augmented generation (VisRAG), and agentic, causal, and embodied AI. Food understanding remains uniquely challenging for AI due to long-tail and culturally diverse food categories, intra-class variability, contextual complexity, and the need for structured meal understanding. Modern systems are moving beyond isolated recognition toward integrated, decision-oriented pipelines combining detection, ingredient understanding, portion and nutrition estimation, and temporal modelling of cooking and eating. Recent advances in Large Language Models (LLMs), VLMs, and multimodal foundation models, coupled with validated nutritional knowledge bases such as food knowledge graphs, increasingly support retrieval-grounded and agentic systems for personalized, explainable, and safety-aware decision support. Inspired by Personal Health Navigator architectures, next-generation systems integrate multimodal signals through closed-loop reasoning and coordinated AI agents. MADIMA2026 therefore focuses on the "Multimodalities of Food": integrating images, video, text, sensors, conversational interactions, and structured nutritional and biomedical information to advance food understanding, personalized recommendations, assistive technologies, embodied AI, and decision-grade systems. Particular emphasis is placed on robustness, inclusiveness, interoperability, scalability, and clinically meaningful evaluation.

Call for Papers

Guidelines for submitting and presenting original research at MADIMA 2026.

Topics of Interest

We invite submissions presenting original research, practical applications, datasets, benchmarks, and perspectives on multimodal artificial intelligence for dietary intelligence, management, and assessment. Topics of interest include, but are not limited to:

Supervised, semi-supervised, weakly supervised, unsupervised, and self-supervised learning for food recognition and understanding
Fine-grained food recognition, transfer learning, few-shot learning, long-tailed learning, and learning with noisy labels
Food detection, localization, segmentation, tracking, and instance counting using conventional and foundation models
Large language models, vision-language models, and multimodal foundation models for recipe understanding, ingredient recognition, food composition analysis, nutritional content estimation, and database grounding
Monocular and stereo depth estimation, three-dimensional reconstruction, and point cloud analysis for food volume and portion size estimation
Augmented and virtual reality for interactive food analysis, dietary logging, portion estimation, and nutrition education
Trustworthy and deployable multimodal AI, including explainability, robustness, fairness, privacy, federated learning, and efficient edge deployment

Multimodal food representation learning across images, video, text, speech, depth, wearable devices, and other sensor data
Out-of-distribution food detection, anomaly detection, open-set recognition, open-world recognition, and uncertainty estimation
Generative models for food image and video synthesis, data augmentation, counterfactual generation, and multimodal content generation
Multimodal dietary intake assessment using food images, videos, speech, text, barcodes, wearable devices, and mobile or ambient sensors
Personalized dietary management, including meal planning, food recommendation, intake monitoring, adaptive feedback, and behavioral support
Temporal and contextual modeling of dietary behavior, meal patterns, food diaries, cuisine diversity, and cultural factors
Datasets, benchmarks, evaluation protocols, and metrics for multimodal food understanding and dietary assessment

Important Dates

Archival track

Full paper submission	September 22, 2026
Notification of acceptance	September 28, 2026
Camera-ready submission	October 4, 2026

Non-archival track

Extended abstract submission	October 18, 2026
Notification of acceptance	November 2, 2026

Nectar track

Submission	October 18, 2026
Notification of acceptance	November 2, 2026

Venue

Osaka International Convention Center (Grand Cube Osaka)

Nakanoshima, Kita-ku, Osaka, Japan

Held in conjunction with ACCV 2026 (Asian Conference on Computer Vision), December 14–18, 2026 - workshops and tutorials on Dec 14–15, main conference Dec 16–18. MADIMA 2026's exact day will be confirmed once ACCV publishes its detailed schedule.

Manuscript Submission

MADIMA 2026 will be held in conjunction with ACCV 2026 (Asian Conference on Computer Vision), with three submission tracks: an archival track (peer-reviewed, included in the proceedings), a non-archival track (extended abstracts, not included in the proceedings), and a nectar track (previously published work, not included in the proceedings).

Submit your paper on OpenReview

Submission Guidelines

Submission and Review

The archival track is peer-reviewed; accepted papers are included in the workshop proceedings. The non-archival and nectar tracks are not peer-reviewed for inclusion in the proceedings. A Springer special issue with selected extended workshop papers is planned.

Paper Format and Length

Papers must be prepared using the official Springer LNCS style files. Authors must download and use the official ACCV 2026 LaTeX template for the main paper.
Archival track papers are limited to 14 pages, including all figures and tables. An unlimited number of additional pages containing only references is allowed.
Submissions must be in PDF format, fully anonymized, and follow the formatting instructions exactly; non-compliant papers or papers that reveal the authors’ identities may be rejected without review.

Download the ACCV 2026 LaTeX template

Double-Blind Review

Review for the archival track is double-blind. Authors must take reasonable care not to reveal their identity in the paper or supplementary material.
Omit acknowledgements from the submitted version, and avoid institutional or lab names, grant numbers, and other identifying information.
Do not include author-identifying information in videos or supplementary files, and avoid links to personal, project, or institutional webpages that reveal authorship.
If you cite your own closely related work, do so in a way that preserves anonymity. Papers that do not follow these anonymity requirements may be rejected without review.

Originality and Dual Submission

Submitted papers must be original and must not be under review at another archival venue during the review period, nor substantially overlap with prior or concurrent published work.
Posting or updating a preprint on arXiv or a similar non-peer-reviewed server during the review period is permitted, provided that anonymity is preserved and the double-blind process is not compromised.
Plagiarism detection and dual-submission checks may be used; papers found to violate these policies may be rejected and may be subject to further action.

AI/LLM Use and Submission Integrity

Submissions must reflect the authors’ own original scholarly contribution. Large Language Models or similar tools must not be used to generate a submission, replace the authors’ intellectual contribution, or fabricate or alter results, citations, or claims.
Limited use of such tools for language editing or formatting assistance is acceptable, provided authors remain fully responsible for all content and the use does not compromise anonymity, originality, or any other guideline.
Submissions must not contain hidden prompts, invisible text, or other concealed content intended to influence reviewers or automated systems. Suspicious or manipulative content may result in desk rejection.

Ethical Responsibilities

Authors are expected to consider the broader impact of their work and to conduct research responsibly, addressing issues such as privacy, fairness, bias, security, or potential misuse where relevant.
If the work uses human-derived, personal, or sensitive data, the paper should describe how the data were collected and handled, and indicate whether ethics approval or exemption was obtained and informed consent secured, where applicable.
Submissions raising significant ethical concerns may undergo additional review.

Adapted from the ACCV 2026 Author Guidelines.

Supplementary Materials

By the submission deadline, authors may optionally submit additional material that was ready at the time of paper submission but could not be included due to constraints of format or space.
Supplementary material may include videos, proofs, additional figures or tables, or more detailed analysis of experiments presented in the paper. There is no page limit for supplementary materials, but only one file with a maximum size of 50MB is allowed.
We encourage (if possible) authors to upload their code as part of their supplementary material to help reviewers assess the quality of the work.