Imgflip & Douban Meme Dataset

We present the first large-scale bilingual affective meme benchmark, designed to bridge the "pixel-to-emotion" gap in multimodal understanding. Comprising over 92,000 samples collected from culturally distinct platforms—Douban (Chinese) and Imgflip (English)—this dataset is unique in providing fine-grained facial emotion annotations that capture the non-literal emotional stances central to internet memes. By treating facial affect as a primary semantic anchor, this benchmark enables researchers to move beyond surface-level text-image alignment and develop robust models for emotion-aware meme retrieval and cross-cultural affective analysis.