List your company

Knowledge-based Visual Question Answering on INFOSEEK Unseen Question

46.5AccuracyDBAgent (SFT: InfoSeek)
6.77217.08627.437.714Dec 15, 2025Jan 15, 2026Feb 16, 2026Mar 19, 2026Apr 20, 2026May 21, 2026Jun 22, 2026

Evaluation results

MethodLinks
DBAgent (SFT: InfoSeek)LLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B + bge-m32026.0446.5———
DBAgent (SFT: Mixed)LLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B + bge-m32026.0443.6———
ReflectiVALLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B2026.0443.5———
VLM-PRFLLM=InternVL3-8B, Retriever=EVA-CLIP-8B2026.0443.5———
VLM-PRF w/ RLLLM/MLLM=InternVL3-8B, Retriever=EVA-CLIP-8B, Feature=V+T2026.0543.5———
CgRAGLLM/MLLM=InternVL3-8B, Retriever=EVA-CLIP-8B, Feature=V+T2026.0543.5———
VLM-PRFLLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B2026.0443.3———
DBAgent (SFT: E-VQA)LLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B + bge-m32026.0443.1———
CgRAGLLM/MLLM=LLaVA-NeXT-8B, Retriever=EVA-CLIP-8B, Feature=V+T2026.0541.9———
CgRAGLLM/MLLM=LLaVA-NeXT-7B, Retriever=EVA-CLIP-8B, Feature=V+T2026.0541.7———
CgRAGLLM/MLLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B, Feature=V+T2026.0541.3———
CgRAGLLM/MLLM=Qwen2-VL-7B, Retriever=EVA-CLIP-8B, Feature=V+T2026.0540.8———
ReflectVALLM=LLaMA-3.1-8B, Retriever=EVA-CLIP-8B2026.0440.4———
VLM-PRFLLM/MLLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B, Feature=V+T2026.0540———
IBA-LLaVABackbone=Llama-3.1-8B, Category=Our proposed IBA2026.0637.838.713.545.9
IBA-QwenBackbone=Qwen-2.5-VL, Category=Our proposed IBA2026.0637.234.57.847.5
MMKB-RAGLLM/MLLM=Qwen2-VL-7B, Retriever=EVA-CLIP-8B, Feature=V2026.0536.4———
ReflectiVABackbone=Llama-3.1-8B, Category=Retrieve Augmented Models requiring Fine-tuning2026.0636.429.710.445.6
2Stage-QwenBackbone=Qwen-2.5-VL, Category=Zero-shot with Re-rank2026.0635.98.4049
1Stage-QwenBackbone=Qwen-2.5-VL, Category=Zero-shot with Re-rank2026.0634.632.71.544.8
Article-qwenBackbone=Qwen-2.5-VL, Category=Zero-shot with Retrieval2026.0634.514.30.246.8
MMhops-R1Model Size=7B2025.1233.8———
EchoSightBackbone=Llama-3.1-8B, Category=Retrieve Augmented Models requiring Fine-tuning2026.0633.824.912.341.7
Wiki-LLaVA2025.1230.1———
Wiki-LLaVALLM=LLaVA-v1.5-7B, Retriever=CLIP ViT-L/14 + Contriever2026.0430.1———
Wiki-LLaVALLM/MLLM=Vicuna-7B, Retriever=CLIP ViT-L/14 + Contriever, Feature=T2026.0530.1———
Wiki-LLaVALLM/MLLM=LLaMA-3.1-8B, Retriever=CLIP ViT-L/14 + Contriever, Feature=T2026.0528.6———
ReflectiVALLM/MLLM=LLaMA-3.1-8B, Retriever=EVA-CLIP-8B, Feature=V2026.0528.6———
2Stage-llavaBackbone=Llama-3.1-8B, Category=Zero-shot with Re-rank2026.0627.26.90.938.5
Para-QwenBackbone=Qwen-2.5-VL, Category=Zero-shot base models2026.0625.510.9035.5
RORA-VLMLLM=LLaVA-v1.5-7B, Retriever=CLIP ViT-L/142026.0425.1———
RORA-VLMLLM/MLLM=Vicuna-7B, Retriever=CLIP + Google Search, Feature=V2026.0525.1———
DBAgent (w/o SFT)LLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B + bge-m32026.0424.9———
LLaVA-NeXT-8BMode=Zero-shot MLLMs2026.0523.7———
CLIP-FiD2025.1223.3———
LLaVA-NeXT-7BMode=Zero-shot MLLMs2026.0523.2———
Qwen2.5-VL-7B (Base)LLM=Qwen2.5-VL-7B (Base), Retriever=–2026.0422.8———
Qwen2.5-VL-7BMode=Zero-shot MLLMs2026.0522.8———
CLIP-PaLM2025.1222.7———
Qwen2.5-VL-3BMode=Zero-shot MLLMs2026.0520.2———
Article-llavaBackbone=Llama-3.1-8B, Category=Zero-shot with Retrieval2026.0618.40026.4
EchoSightLLM/MLLM=LLaMA-3.1-8B, Retriever=EVA-CLIP-8B, Feature=V2026.0518———
Qwen2-VL-7BMode=Zero-shot MLLMs2026.0515.4———
GPT-4VLLM=GPT-4V, Retriever=–2026.0415———
GPT-4VMode=Zero-shot MLLMs2026.0515———
BLIP-2LLM=Flan-T5XL, Retriever=–2026.0412.7———
BLIP-2LLM/MLLM=Flan-T5XL, Mode=Zero-shot MLLMs2026.0512.7———
1Stage-llavaBackbone=Llama-3.1-8B, Category=Zero-shot with Re-rank2026.0610.50015
LLaVA-v1.5LLM=Vicuna-7B, Retriever=–2026.049.6———
LLaVA-1.5LLM/MLLM=Vicuna-7B, Mode=Zero-shot MLLMs2026.059.6———
Para-LlavaBackbone=Llama-3.1-8B, Category=Zero-shot base models2026.0690.90.512.4
InstructBLIPLLM=Flan-T5XL, Retriever=–2026.048.9———
InstructBLIPLLM/MLLM=Flan-T5XL, Mode=Zero-shot MLLMs2026.058.9———
LLaVA-1.5LLM/MLLM=LLaMA-3.1-8B, Mode=Zero-shot MLLMs2026.058.3———