Zero-shot Video Emotional Description Generation Based on Counterfactual Emotional Causal Reasoning

Key Points

• The aim is to develop a zero-shot method for generating emotional video descriptions without relying on large annotated datasets.
• Proposed a causal reasoning framework to enhance existing visual-language models.
• Generated pseudo-labels for videos covering global content, emotions, and facial expressions.
• Implemented direct effect modeling to establish causal relationships between visual and linguistic aspects.
• Demonstrated improved performance on emotional description generation metrics compared to current methods.
• Verified the effectiveness of the proposed framework in both in-domain and out-of-domain experiments.