List your company
ResearchFormerly WizWand

Visual Question Answering (Multi-choice) on A-OKVQA (test)

87.2AccuracyHinD-CoT-Know
38.21650.93363.6576.367Nov 15, 2022May 20, 2023Nov 23, 2023May 27, 2024Nov 30, 2024Jun 4, 2025Dec 8, 2025
Updated 7mo ago

Evaluation results

MethodLinks
HinD-CoT-KnowModel Size=7B2025.1187.2
HinD-KnowModel Size=7B2025.1185.9
ReAuSE2025.1185
BLIP-2Training data=129M image-text pairs2023.0480.2
LLaVA-1.5Model Size=7B2025.1177.1
InstructBLIPLLM Backbone=FlanT5-XXL, Evaluation Protocol=finetuning2023.0576.7
QACap2025.1176.7
Prophet2025.1176.4
BLIP-2LLM Backbone=FlanT5-XXL, Evaluation Protocol=finetuning2023.0576.2
Previous SOTAEvaluation Protocol=finetuning2023.0573.6
InstructBLIPLLM Backbone=Vicuna-7B, Evaluation Protocol=finetuning2023.0573.4
PromptCap2025.1173.2
PromptCap + GPT-3Example retrieval=CLIP (ViT-L/14)2022.1173.1
InstructBLIPModel Size=7B, Training=Fine-tuned (FT)2025.1173
SAVEBackbone=LLaVA-1.6, Layer=20, Steering Strength=52025.1270.04
BLIP-2LLM Backbone=Vicuna-7B, Evaluation Protocol=finetuning2023.0569
MiniGPT-4 (Finetune Vicuna)Training data=5M image-text pairs, Unfreeze LLM=using LoRA2023.0467.2
BaseBackbone=LLaVA-1.62025.1266.89
SAVEBackbone=LLaVA-NeXT, Layer=24, Steering Strength=102025.1266.81
BaseBackbone=LLaVA-NeXT2025.1265.58
GPT-22025.1160.3
MiniGPT-4Training data=5M image-text pairs2023.0458.2
GPV-22022.1153.7
ClipCap2022.1143.8
KRISP2022.1142.2
LXMERT2022.1141.6
ViLBERT2022.1141.5
Pythia2022.1140.1