EVATok: Adaptive Length Video Tokenization for Efficient Visual Autoregressive Generation

Benchmark Model Rank Results
video-generation-on-ucf-101EVATok (327M)–FVD16: 62
video-generation-on-ucf-101EVATok (633M)–FVD16: 48
video-prediction-on-kinetics-600-12-frames-64x64EVATok (327M)–FVD: 4.6
video-prediction-on-kinetics-600-12-frames-64x64EVATok (633M)–FVD: 4.0