CN-VEFD: A Visual Emotion Feature Dataset for Chinese ESG Reports

CN-VEFD is a large-scale dataset for visual emotion analytics in ESG reporting. It compiles 13,481 ESG reports from Chinese listed companies (2006–2023) and programmatically extracts 399,321 images. Each image is annotated with 66 structured visual emotion features spanning three categories: color emotion (e.g., positive/negative/neutral color ratios, harmony and rhythm metrics), composition emotion (e.g., saliency-based visual balance, whitespace ratio, edge/line/circle structure statistics), and facial emotion (e.g., image-level emotion distribution and polarity derived from face detection and FER models). The dataset is produced via automated pipelines combining color analysis, saliency detection, geometric analysis, and facial emotion recognition (MTCNN + FERplus), with deduplication, quality filtering, normalization, and manual validation on a subset to ensure reliability. CN-VEFD enables research in visual analytics, emotion recognition, ESG disclosure evaluation, and behavioral finance, supporting studies on how visual elements relate to governance quality, disclosure effectiveness, and market responses.