Divide, then Ground: Adapting Frame Selection to Query Types for Long-Form Video Understanding

Researchers propose DIG, a training-free framework that improves long-form video understanding by adapting frame selection strategies based on query types. The system uses uniform sampling for global queries and specialized selection for localized queries, achieving better performance than existing methods while scaling to 256 input frames.

• →DIG framework distinguishes between global and localized queries to optimize frame selection in video analysis.
• →Uniform sampling proves effective for global queries while localized queries require query-aware selection methods.
• →The training-free approach reduces computational overhead compared to complex search mechanisms.
• →DIG consistently outperforms existing baselines across three long-form video understanding benchmarks.
• →The framework successfully scales to process 256 input frames while maintaining robust performance improvements.