List your company

Video Understanding on MVBench, EgoSchema, MLVU, LongVideoBench, VideoMME

56.3Overall ScoreVanilla
38.82843.36447.952.436Mar 23, 2026

Evaluation results

MethodLinks
VanillaPrefill (ms)=701.2 (1×), TTFT (ms)=1104.7, Throughput (token/s)=27.62026.0356.3——
LLaVA-OV-7BRetention Ratio=100%2026.0356.3100—
OursRetention Ratio=15%/7.5%2026.0356.199.6—
HoliTomRetention Ratio=15%/7.5%2026.035699.5—
OursRetention Ratio=10%/5%2026.0355.498.4—
HoliTomRetention Ratio=10%/5%2026.0355.398.3—
Ours (w/o M)Retention Ratio=10%2026.0355.398.3—
FastVIDRetention Ratio=15%2026.0354.496.6—
FastVIDRetention Ratio=10%2026.0354.296.2—
VisionZipRetention Ratio=15%2026.035495.9—
DyToKRetention Ratio=15%2026.035495.9—
OursRetention Ratio=5%/2.5%2026.0353.795.4—
Ours (w/o M)Retention Ratio=5%2026.0353.494.9—
Qwen2.5-VLRetention Ratio=100%2026.0353.4—100
HoliTomRetention Ratio=5%/2.5%2026.0352.994—
DyToKRetention Ratio=10%2026.0352.693.5—
FastVIDRetention Ratio=5%2026.0351.591.5—
VisionZipRetention Ratio=10%2026.0351.391.2—
FsatVRetention Ratio=100%/10%2026.0351.190.8—
Unified Spatiotemporal Token CompressionProcess (ms)=74.0, Prefill (ms)=31.3 (22×), TTFT (ms)=473.8, Throughput (token/s)=35.72026.0350.7——
OursRetention Ratio=2%/1%2026.0350.790.1—
Ours (w/o M)Retention Ratio=2%2026.0350.489.5—
HoliTomProcess (ms)=88.7, Prefill (ms)=40.5 (17×), TTFT (ms)=497.4, Throughput (token/s)=34.32026.0349.4——
HoliTomRetention Ratio=2%/1%2026.0349.487.7—
DyToKRetention Ratio=5%2026.0348.986.8—
FastVRetention Ratio=100%/5%2026.0347.784.7—
Ours (w/o M)Retention Ratio=1%2026.0347.484.2—
OursRetention Ratio=1%/0.5%2026.0347.384.1—
FastVidProcess (ms)=8.6, Prefill (ms)=43.1 (16×), TTFT (ms)=446.2, Throughput (token/s)=33.82026.0346.9——
FastVIDRetention Ratio=2%2026.0346.983.3—
HoliTomRetention Ratio=1%/0.5%2026.0346.782.9—
VisionZipProcess (ms)=35.8, Prefill (ms)=42.1 (17×), TTFT (ms)=451.6, Throughput (token/s)=35.42026.0345.8——
VisionZipRetention Ratio=5%2026.0345.881.4—
FastVRetention Ratio=100%/2%2026.0345.681—
Ours(w/o M)Retention Ratio=2%2026.0345.6—85.5
OursRetention Ratio=2%2026.0345.5—85.3
FastVRetention Ratio=100%/1%2026.0344.278.5—
FastVIDRetention Ratio=2%2026.0344—82.5
FastVIDRetention Ratio=1%2026.0343.777.7—
DyToKRetention Ratio=2%2026.0343.577.3—
VisionZipRetention Ratio=2%2026.0342.6—79.8
OursRetention Ratio=1%2026.0342.6—79.8
VisionZipRetention Ratio=2%2026.0342.475.3—
Ours(w/o M)Retention Ratio=1%2026.0342.4—79.4
DyToKRetention Ratio=1%2026.0341.373.3—
FastVIDRetention Ratio=1%2026.0341.1—76.9
VisionZipRetention Ratio=1%2026.0340.672.1—
VisionZipRetention Ratio=1%2026.0339.5—74