EarlyTom: Early Token Compression Completes Fast Video Understanding
Open paper
Benchmark
Model
Rank
Results
zero-shot-video-question-answer-on-egoschema-fullset
EarlyTom (LLaVA-OneVision-7B)
#31
Accuracy: 60.6
video-question-answering-on-mvbench
EarlyTom (LLaVA-OneVision-7B)
#76
Avg.: 57.8
Rank counts only results with a code link.