List your company

Robotic Manipulation on VIMA-Bench 1.0 (test)

81.5L1 ScoreVIMA
45.51654.85864.273.542Dec 28, 2023May 4, 2024Sep 10, 2024Jan 17, 2025May 25, 2025Oct 1, 2025Feb 7, 2026

Evaluation results

MethodLinks
VIMAInput Representation=Object tokens (cropped images and bounding boxes)2023.1281.581.578.748.672.6—
iSTARVariant=Full model2026.0278.679.278.344.3——
iSTARVariant=w/o Concept Alignment2026.0277.877.57740——
iSTARVariant=w/o Order Entropy2026.0276.477.676.342.7——
iSTARVariant=w/o Graph2026.0274.87674.937.7——
VIMAModel size=Large2026.0273.974.57259——
VIMAModel size=Small2026.0272.473.871.135——
iSTARVariant=Stacked VIMA2026.0272.473.672.434——
UNIFIED-IO 2 XXLModel Scale=XXL2023.1271.370.46815.556.3—
UNIFIED-IO 2 XLModel Scale=XL2023.1270.369.864.513.154.2—
UNIFIED-IO 2 LModel Scale=L2023.1266.963.857.512.650.2—
VIMA-GatoInput Representation=Image patch tokens (ViT)2023.125753.945.613.542.5—
VIMA-FlamingoInput Representation=Image patch tokens (ViT) + downsampled by perceiver module2023.1247.44640.712.136.6—
VIMA-GPTInput Representation=Single image token (ViT)2023.1246.946.942.212.137—
LLaRA-dist#Data=1K2025.03—————53.9
LLaRA-dist#Data=10K2025.03—————83.4
LLaRA-dist#Data=100K2025.03—————89.5
LLaRA-sft#Data=1K2025.03—————49.6
LLaRA-sft#Data=10K2025.03—————82.3
LLaRA-sft#Data=100K2025.03—————88.5
LLaRA-vocab#Data=1K2025.03—————50.8
LLaRA-vocab#Data=10K2025.03—————81
LLaRA-vocab#Data=100K2025.03—————87
RT-2#Data=1K2025.03—————1.9
RT-2#Data=10K2025.03—————21.9
RT-2#Data=100K2025.03—————73.1