Automatic Funny Scene Extraction from Long-form Cinematic Videos
The multimodal system analyzes audio and text to find humor, improving scene detection by 18.3%.
Deep Dive
Researchers Sibendu Paul, Haotian Jiang, and Caren Chen developed an end-to-end AI system for automatically identifying humorous scenes in long movies. Key innovations include a novel multimodal scene segmentation approach and a humor tagging framework. The pipeline achieves an 87% accuracy rate for extracting intended funny clips and an 18.3% AP improvement in scene detection. It can streamline content creation for streaming platforms by generating engaging previews and snackable clips.
Why It Matters
Automates the creation of viral movie clips and trailers, boosting engagement for streaming services and content creators.