List your company
ResearchFormerly WizWand

Multimodal Perception Assessment on MME Perception

1,702.17MME-PQwen2-VL-7B
1,032.23321,206.15911,380.0851,554.0109Mar 9, 2025May 21, 2025Aug 3, 2025Oct 16, 2025Dec 29, 2025Mar 13, 2026May 26, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen2-VL-7BBackbone=Qwen2-VL-7B2026.041,702.17
HalluVL-DPOBackbone=Qwen2-VL-7B2026.041,696.96
AD-Loop#Params=7B, Capabilities=Und. and Gen.2026.021,696
HalluVL-DPOwBackbone=Qwen2-VL-7B2026.041,691.17
BAGEL#Params=7B, Capabilities=Und. and Gen.2026.021,687
VCDBackbone=Qwen2-VL-7B2026.041,682.13
CoVFTBackbone=LLaVA-1.5-13B2026.031,590.6
ShareGPT4V# Params=7B, Resolution=3362025.031,567.4
Janus-Pro#Params=7B, Capabilities=Und. and Gen.2026.021,567.1
FreezeBackbone=LLaVA-1.5-13B2026.031,557.6
Full fine-tuningBackbone=LLaVA-1.5-13B2026.031,555.4
BitFitBackbone=LLaVA-1.5-13B2026.031,552.8
LLaVA-1.5Backbone=LLaVA-1.5-13B2026.031,531.3
CoVFTBackbone=LLaVA-1.5-7B2026.031,525.2
SemHiTok# Params=7B, Resolution=3842025.031,512.8
LLaVA-v1.5# Params=7B, Resolution=3362025.031,510.7
LLaVA-v1.5#Params=7B, Capabilities=Und. Only2026.021,510.7
LLaVA-1.5Backbone=LLaVA-1.5-7B2026.031,510.7
Full fine-tuningBackbone=LLaVA-1.5-7B2026.031,510.2
VisNecBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,505.5
BitFitBackbone=LLaVA-1.5-7B2026.031,496.8
LoRABackbone=LLaVA-1.5-7B2026.031,494.5
LLaVA-1.5-7BBackbone=LLaVA-1.5-7B2026.041,493.23
TokLIP# Params=7B, Resolution=3842025.031,488.4
Qwen-Vl-Chat# Params=7B, Resolution=4482025.031,487.5
Qwen-VL-Chat#Params=7B, Capabilities=Und. Only2026.021,487.5
XMASSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,485.7
Full-DataSampling Ratio=100% (665K), Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,476.9
FreezeBackbone=LLaVA-1.5-7B2026.031,473.7
PreSelSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,457.7
VPTBackbone=LLaVA-1.5-7B2026.031,452.1
SemHiTok# Params=7B, Resolution=2562025.031,449
Liquid#Params=8B, Capabilities=Und. and Gen.2026.021,448
Janus-Pro-1B# Params=1.5B, Resolution=3842025.031,444
OFASampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,443.4
ICONSSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,435.6
SVPTBackbone=LLaVA-1.5-7B2026.031,431.7
AntidoteBackbone=LLaVA-1.5-7B, Hyperparameter Configuration=HP-orig2026.041,422.77
CoIDOSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,419.5
COINCIDESampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,414.9
MMaDA#Params=8B, Capabilities=Und. and Gen.2026.021,410.7
EL2NSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,405.2
VILA-U# Params=7B, Resolution=3842025.031,401.8
RandomSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,397.5
TypiClustSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,396.2
VCDBackbone=LLaVA-1.5-7B2026.041,389.79
SynerGen-VL# Params=2.4B, Resolution=5122025.031,381
CLIP-ScoreSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,380.3
TokenFlow-L# Params=13B, Resolution=2562025.031,365.4
HalluVL-DPOBackbone=LLaVA-1.5-7B2026.041,359.95
Self-FilterSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,356.5
TokenFlow-B# Params=13B, Resolution=2242025.031,353.6
Janus# Params=1.5B, Resolution=3842025.031,338
VILA-U# Params=7B, Resolution=2562025.031,336.2
LLaVA-Phi# Params=2.7B, Resolution=2562025.031,335.1
HalluVL-DPOwBackbone=LLaVA-1.5-7B2026.041,314.8
IFDSampling Ratio=15%, Backbone=LLaVA-v1.5-7B, Training Dataset=LLaVA-665K2026.051,307.2
CoIDOBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,298.8
Self-FilterBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,262.2
EL2NBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,253.8
ICONSBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,252.5
Emu3#Params=8B, Capabilities=Und. and Gen.2026.021,244
EMU3# Params=8B, Resolution=5122025.031,243.8
RandomBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,238.6
Full-Data(186K)Backbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=100%2026.031,238.1
AntidoteBackbone=LLaVA-1.5-7B, Hyperparameter Configuration=HP-search2026.041,220.05
PreSelBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,218.8
InstructBLIP#Params=13B, Capabilities=Und. Only2026.021,212.8
IFDBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,210.9
TypiClustBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,194.1
COINCIDEBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,184.4
MAR# Params=1.5B, Resolution=5122025.031,155
XMASBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,151.2
Liquid# Params=7B, Resolution=5122025.031,119.3
Show-o# Params=1.5B, Resolution=2562025.031,097.2
Show-o#Params=1.3B, Capabilities=Und. and Gen.2026.021,097.2
CLIP-ScoreBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.031,058