Pretraining Frame Preservation in Autoregressive Video Memory Compression
IntermediateLvmin Zhang, Shengqu Cai et al.Dec 29arXiv
The paper teaches a video model to squeeze long video history into a tiny memory while still keeping sharp details in single frames.
#autoregressive video generation#video memory compression#frame retrieval pretraining