List your company
ResearchFormerly WizWand

Question Answering on EgoSchema

58.4AccuracyQwen2.5-Omni
10.97623.28835.647.912Jan 15, 2026Feb 5, 2026Feb 26, 2026Mar 19, 2026Apr 9, 2026Apr 30, 2026May 21, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen2.5-Omnievaluation_protocol=spoken questions2026.0158.4
LLaVA-VideoBase Model=LLaVA-Video-7B, Input Frames=64, Mode=Upper Bound (Full Performance)2026.0557.3
ST-SimDiffBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=50%2026.0557.3
FasterVLMBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=50%2026.0556.2
ST-SimDiffBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=30%2026.0556
FrameFusionBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=50%2026.0555.8
ROMAevaluation_protocol=spoken questions2026.0155.4
MiniCPM-oevaluation_protocol=spoken questions2026.0155.2
ROMAevaluation_protocol=spoken questions, wpos=42026.0154.8
FastVBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=50%2026.0554.7
PruMergeBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=50%2026.0554.6
VisionZipBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=50%2026.0554.2
ROMAevaluation_protocol=spoken questions, K=12026.0154
VisionZipBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=30%2026.0553
FrameFusionBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=30%2026.0553
ROMAevaluation_protocol=spoken questions, wpos=22026.0152.6
FasterVLMBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=30%2026.0552.6
FastVBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=30%2026.0551.3
PruMergeBase Model=LLaVA-Video-7B, Input Frames=64, Token Retain Ratio (r)=30%2026.0550.9
ROMAevaluation_protocol=spoken questions, ablation=Mixed Training2026.0150.2
VITA-1.5evaluation_protocol=spoken questions2026.0145.4
ROMAevaluation_protocol=spoken questions, ablation=without speak head2026.0112.8