List your company
ResearchFormerly WizWand

Image Understanding on MMStar

65.1ScoreGPT-4o-2024-11-20
34.62842.53950.4558.361Oct 27, 2025Dec 2, 2025Jan 7, 2026Feb 13, 2026Mar 21, 2026Apr 26, 2026Jun 2, 2026
Updated 5mo ago

Evaluation results

MethodLinks
GPT-4o-2024-11-20Version=2024-11-202026.0365.1
SpatialSSRLBackbone=Qwen2.5-VL-7B2026.0463.5
ROBOALIGN (Ours)Training=SFT + RL2026.0362.8
SpatialEvoBackbone=Qwen2.5-VL-7B2026.0462.5
DCRL2026.0662.5
Top-k RoutingExpert skipping ratio (ρ)=0, Number of experts (k)=8, Backbone=InternVL-3.5-30B-A3B-HF2025.1162.49
ROBOALIGN w/o RLTraining=SFT only2026.0362.47
MoDESExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1162.06
VeBrain-8BParameters=8B2026.0361.9
DiEPExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1161.82
BaselineBackbone=Qwen2.5-VL-7B2026.0461.6
MoDESExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1161.46
MC-MoEExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1161.45
MoDESExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1161.2
NAEEExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1161.18
ViLaSRBackbone=Qwen2.5-VL-7B2026.0460.8
Top-k RoutingExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1160.3
Qwen2.5-VL-7B-InsParameters=7B, Variant=Instruct2026.0360.3
Qwen3-VL-8BModel Size=8B2026.0659.8
NAEEExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1158.82
DiEPExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1158.42
MC-MoEExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1157.65
DiEPExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1157.21
SpatialSSRLBackbone=Qwen2.5-VL-3B2026.0456.5
MC-MoEExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1156.14
SpatialEvoBackbone=Qwen2.5-VL-3B2026.0455.2
SpaceRBackbone=Qwen2.5-VL-7B2026.0454.9
BaselineBackbone=Qwen2.5-VL-3B2026.0454.6
Cosmos-Reason1-7BParameters=7B2026.0354.4
CircleRoPEBackbone=Qwen3-VL-8B-Instruct2025.1054.4
MRoPE-IBackbone=Qwen3-VL-8B-Instruct2025.1053.47
VideoRoPEBackbone=Qwen3-VL-8B-Instruct2025.1053.33
MHRoPEBackbone=Qwen3-VL-8B-Instruct2025.1053.2
HoPEBackbone=Qwen3-VL-8B-Instruct2025.1052.86
MRoPEBackbone=Qwen3-VL-8B-Instruct2025.1052.53
Vanilla RoPEBackbone=Qwen3-VL-8B-Instruct2025.1051.88
MRoPE-I2025.1051.13
HoPE2025.1050.33
NAEEExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1150.14
MRoPE2025.1049.93
VideoRoPE2025.1049.6
MHRoPE2025.1049.6
Vanilla RoPE2025.1049.13
CircleRoPE2025.1047
Top-k RoutingExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1146.69
SpatialLadderBackbone=Qwen2.5-VL-3B2026.0445.8
MRoPE-IBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MRoPE-I2025.1045
SpaceRBackbone=Qwen2.5-VL-3B2026.0444.9
CircleRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=CircleRoPE2025.1043.73
MHRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MHRoPE2025.1043.73
Vanilla RoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=Vanilla RoPE2025.1043.67
HoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=HoPE2025.1043.53
MRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MRoPE2025.1042.87
Top-k RoutingExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1142.27
VideoRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=VideoRoPE2025.1041.87
RoboBrain2.0-7BVersion=2.0, Parameters=7B2026.0335.8