Reasoning with Memory: A Temporal Granularity-Adaptive Framework for Training-Free Long Video Understanding
Open paper
Benchmark
Model
Rank
Results
video-question-answering-on-lvbench
ReMem (LLaVA-Video-7B-Qwen2, 64 frames)
–
Accuracy: 54.5
video-question-answering-on-mlvu
ReMem (Qwen3-VL-8B-Instruct, 32 frames)
–
Accuracy: 77.6
Rank counts only results with a code link.