List your company
ResearchFormerly WizWand

Spatial Understanding on COCO-QA v1.5 (Spatial)

71.51COCO-QA_Spat AccuracyLLaVAflickr SiTe-ratio
46.75853.18459.6166.036Dec 7, 2025
Updated 8mo ago

Evaluation results

MethodLinks
LLaVAflickr SiTe-ratioTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=SiTe-ratio, Pretraining Dataset=Flickr30K2025.1271.51
LLaVAflickr SiTe-randTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=SiTe-rand, Pretraining Dataset=Flickr30K2025.1271.42
LLaVASiTe-ratioTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=SiTe-ratio, Pretraining Dataset=558K2025.1270.06
LLaVAflickr BaselineTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=Baseline, Pretraining Dataset=Flickr30K2025.1268.96
LLaVA_1K_SiTe-randTraining Stage=Supervised Fine-tuning, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=SiTe-rand, SFT Sample Count=1K2025.1268.81
LLaVASiTe-randTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=SiTe-rand, Pretraining Dataset=558K2025.1268.75
LLaVA_5K_SiTe-ratioTraining Stage=Supervised Fine-tuning, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=SiTe-ratio, SFT Sample Count=5K2025.1268.37
LLaVARotateTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=Rotate, Pretraining Dataset=558K2025.1268.09
LLaVAVSRTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=VSR, Pretraining Dataset=558K2025.1267.85
LLaVACropTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=Crop, Pretraining Dataset=558K2025.1267.82
LLaVABaselineTraining Stage=Pretraining, Backbone Architecture=LLaVA-v1.5-7B, Data Augmentation Method=Baseline, Pretraining Dataset=558K2025.1267.72
HALVASiTeTraining Stage=Supervised Fine-tuning, Backbone Architecture=HALVA, Data Augmentation Method=SiTe2025.1264.77
Qwen2SiTe-randTraining Stage=Pretraining, Backbone Architecture=LLaVA-Qwen2-1.5B, Data Augmentation Method=SiTe-rand, Pretraining Dataset=558K2025.1262.57
Qwen2SiTe-ratioTraining Stage=Pretraining, Backbone Architecture=LLaVA-Qwen2-1.5B, Data Augmentation Method=SiTe-ratio, Pretraining Dataset=558K2025.1262.52
Qwen2BaselineTraining Stage=Pretraining, Backbone Architecture=LLaVA-Qwen2-1.5B, Data Augmentation Method=Baseline, Pretraining Dataset=558K2025.1262.25
Qwen2flickr SiTe-ratioTraining Stage=Pretraining, Backbone Architecture=LLaVA-Qwen2-1.5B, Data Augmentation Method=SiTe-ratio, Pretraining Dataset=Flickr30K2025.1249.04
Qwen2flickr SiTe-randTraining Stage=Pretraining, Backbone Architecture=LLaVA-Qwen2-1.5B, Data Augmentation Method=SiTe-rand, Pretraining Dataset=Flickr30K2025.1247.71