Zeroshot Video Question Answer
The Zero-Shot Video Question Answering task aims to enable large language models to accurately answer questions about video content without specific training. This task falls under the domain of computer vision and enhances the model's cross-modal understanding capabilities, allowing for immediate analysis and response to unseen video data. It has significant application value, especially in intelligent dialogue systems, video content retrieval, and automatic question answering scenarios.
ActivityNet-QA
IG-VLM
CinePile: A Long Video Question Answering Dataset and Benchmark
EgoSchema (fullset)
BIMBA-LLaVA-Qwen2-7B
EgoSchema (subset)
Tarsier (34B)
IntentQA
IG-VLM
MSRVTT-QA
Flash-VStream
MSVD-QA
Video-LLaVA-7B
MVBench
TS-LLaVA-34B
NExT-GQA
NExT-QA
Tarsier (34B)
STAR Benchmark
VideoChat2
TGIF-QA
PLLaVA
TVQA
FrozenBiLM (with speech)
Video-MME
Gemini 1.5 Pro
Video-MME (w/o subs)
Video-RAG (based on LLaVA-Video)
Zero-shot Video Question Answering on LongVideoBench
Gemini 1.5 Pro