Overview of the CC-MMD 2026 Grand Challenge: Cross-Cultural Misogynistic Meme Detection in Multimodal Memes

Abstract

The Cross-Cultural Misogynistic Meme Detection Grand Challenge, CC-MMD 2026, addresses the problem of identifying misogynistic content in multimodal memes across culturally diverse annotation perspectives. Existing misogyny detection benchmarks have advanced multimodal content moderation, but most assume a single ground-truth label and provide limited insight into how systems behave when harmfulness judgments vary across cultural contexts. CC-MMD introduces a multilingual and multimodal evaluation setting covering Tamil, Malayalam, Chinese, and English memes, with annotations from Indian, Chinese, and Irish cultural perspectives. The challenge defines two tasks: single-culture prediction, which evaluates classification from one selected cultural perspective, and cross-cultural prediction, which requires systems to predict all available cultural labels for each meme. This paper presents the task formulation, dataset construction, annotation design, evaluation protocol, participating systems, and official results. A total of 70 teams participated, with 60 valid submitted runs covering Task A and Task B. Submitted systems included prompt-based vision-language models, parameter-efficient fine-tuning of large multimodal models, dual-encoder fusion architectures, culture-aware modelling strategies, ensemble methods, and lightweight baselines. The results show that strong performance depends on joint modelling of image and text, while cross-cultural prediction additionally benefits from culture-sensitive adaptation, such as culture-specific prompting, multi-head prediction, gated fusion, or explicit cultural grounding. CC-MMD provides a benchmark for studying culturally aware multimodal content moderation and highlights the need to evaluate harmful content detection across culturally situated perspectives rather than relying on a single universal label.