2025 ICML ICML 2025

LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding