List your company

Real-time Visual Perception on OVO-Bench Reactive QA 1.0 (test)

65.1OCRROMA
5.76821.171536.57551.9785Jan 15, 2026

Evaluation results

MethodLinks
ROMAPositive loss weight (wpos)=42026.0165.154.1368.9738.270.361.41
ROMAPositive loss weight (wpos)=22026.0164.4351.3868.9739.3364.3660.87
ROMA2026.0163.0953.2168.139.3369.3158.15
ROMATraining strategy=Mixed Training2026.0163.0955.0563.7937.6461.3955.43
ROMAAggregation layers (K)=12026.0161.4755.0568.139.8965.3560.33
ROMAArchitecture configuration (Speak Head)=Without Speak Head2026.0161.0755.0563.9739.8965.3554.89
Dispider2026.0157.7249.5462.0744.9461.3951.63
Flash-VStream-7B2026.0124.1629.3628.4533.7125.7428.8
MMDuet2026.0113.4211.9314.6611.814.8510.33
VideoLLM-online2026.018.0523.8512.0714.0445.5421.2