List your company

Spatial Reasoning on COCO 2017 (val)

83Alignment AccuracyLLaVA (Group 1 Baseline)
5763.7570.577.25Jan 23, 2026

Evaluation results

MethodLinks
LLaVA (Group 1 Baseline)Model Scale=1.5-13B2026.01831431
LLaVA (Group 2 Baseline)Model Scale=1.5-13B2026.01801429
Rotation TextProtocol=Direct2026.01808987
Emb. TextProtocol=Direct2026.01758381
Rotation TextProtocol=CoT2026.01708682
Rotation TokensProtocol=Direct2026.01708984
Emb. TextProtocol=CoT2026.01679488
Rotation TokensProtocol=CoT2026.01609487
Emb. Tokens (COCO)Protocol=Direct2026.01588175
Emb. Tokens (COCO)Protocol=CoT2026.01589283
Emb. Tokens (ViTPose)Protocol=Direct2026.01588679
Emb. Tokens (ViTPose)Protocol=CoT2026.01588377