List your company
ResearchFormerly WizWand

Video Question Answering on NExT-QA zero-shot

0.845AccuracyVideoLLaMA3
0.17420.348350.52250.69665Apr 5, 2023Oct 16, 2023Apr 27, 2024Nov 7, 2024May 20, 2025Nov 30, 2025Jun 12, 2026
Updated 4mo ago

Evaluation results

MethodLinks
VideoLLaMA3LLM=Qwen2.5-7B, Frames=1fps2025.070.845—
InternVL3.5-REGATELLM=Qwen3-14B, Frames=16, Tokens=2.32B (↓ 41.41%)2025.070.812—
InternVL3.5LLM=Qwen3-14B, Frames=16, Tokens=3.96B2025.070.808—
LLaVA-OneVisionLLM=Qwen2-7B, Frames=322025.070.794—
Original (upper bound)protocol=zero-shot, evaluation model=Qwen3-VL-8B2026.060.7831—
FAMSprotocol=zero-shot, evaluation model=Qwen3-VL-8B, category=Face-only, generative2026.060.7797-0.34
DeepPrivacy2 (face)protocol=zero-shot, evaluation model=Qwen3-VL-8B, category=Face-only, generative2026.060.778-0.51
EgoBlurprotocol=zero-shot, evaluation model=Qwen3-VL-8B, category=Classical2026.060.7756-0.75
LLaVA-NeXT-VideoLLM=Qwen2-32B, Frames=322025.070.773—
FADMprotocol=zero-shot, evaluation model=Qwen3-VL-8B, category=Body-only, generative2026.060.7649-1.82
StructAllprotocol=zero-shot, evaluation model=Qwen3-VL-8B, conditioning=Wan2.1-VACE2026.060.7585-2.46
VideoChat2-REGATELLM=Mistral-7B, Frames=16, Tokens=2.22B (↓ 43.51%)2025.070.755—
StructHumanprotocol=zero-shot, evaluation model=Qwen3-VL-8B, conditioning=Wan2.1-VACE2026.060.7537-2.94
VideoChat2LLM=Mistral-7B, Frames=16, Tokens=3.93B2025.070.752—
Pixelationprotocol=zero-shot, evaluation model=Qwen3-VL-8B, category=Classical2026.060.7376-4.55
DeepPrivacy2 (body)protocol=zero-shot, evaluation model=Qwen3-VL-8B, category=Body-only, generative2026.060.7339-4.92
Blackoutprotocol=zero-shot, evaluation model=Qwen3-VL-8B, category=Classical2026.060.73-5.31
Gaussian blurprotocol=zero-shot, evaluation model=Qwen3-VL-8B, category=Classical2026.060.7293-5.38
VideoLLaMA2LLM=Qwen2-7B, Frames=16, Tokens=83.82M2025.070.708—
VideoLLaMA2-REGATELLM=Qwen2-7B, Frames=16, Tokens=49.27M (↓ 41.22%)2025.070.7—
VILA1.5LLM=Llama-2-40B, Frames=82025.070.679—
SeViLAParams=4.1B, Zero-shot=true2023.040.636—
InternVideoType=Enc-Dec, Params=1.3B, Zero-shot=true2023.040.491—
VicTR (B/16)Params=167M, Zero-shot=true2023.040.455—
X-CLIPType=Enc only, Params=194M, Zero-shot=true2023.040.438—
Just-AskParams=75M, Zero-shot=true2023.040.384—
CaKE-LMParams=2.7B, Zero-shot=true2023.040.349—
RandomZero-shot=true2023.040.2—