Slide Window Video Captioning With Temporal Context Preservation

via “slide-window video captioning with temporal context preservation”

[NeurIPS 2024] An official implementation of "ShareGPT4Video: Improving Video Understanding and Generation with Better Captions"

Unique: Uses sliding window approach with configurable stride to balance temporal context capture against computational cost; generates captions that explicitly model event sequences and transitions rather than treating frames independently

vs others: Produces more semantically coherent captions than frame-by-frame approaches; enables better temporal understanding than single-frame vision models while remaining more efficient than recurrent video encoders