Knowledge-Based Visual Question Answering on E-VQA Single-Hop
59.9AccuracyPMSR
Evaluation Results
| Method | Links | |
|---|---|---|
| PMSRRetriever=SigLIP2-g, Model=Gemini-2.5-Flash2025.08 | 59.9 | |
| OMGMRetriever=EVA-CLIP-8B, Model=GPT-4o, Reranking=true2025.08 | 51.2 | |
| OMGMRetriever=EVA-CLIP-8B, Model=LLaVA-1.5-7B, Reranking=true2025.08 | 50.2 | |
| DBAgent (SFT: E-VQA)LLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B + bge-m32026.04 | 46.5 | |
| PMSRRetriever=SigLIP2-g, Model=Qwen3-VL-8B2025.08 | 46.4 | |
| DBAgent (SFT: Mixed)LLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B + bge-m32026.04 | 46 | |
| DBAgent (SFT: InfoSeek)LLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B + bge-m32026.04 | 45.1 | |
| PMSRRetriever=SigLIP2-g, Model=Qwen3-VL-4B2025.08 | 40.9 | |
| CgRAGLLM/MLLM=InternVL3-8B, Retriever=EVA-CLIP-8B, Feature=V+T2026.05 | 40.4 | |
| VLM-PRFLLM=InternVL3-8B, Retriever=EVA-CLIP-8B2026.04 | 40.1 | |
| Wiki-PRF(w/ RL)Retriever=EVA-CLIP-8B, Model=VLM-PRF-7B2025.08 | 40.1 | |
| VLM-PRF w/ RLLLM/MLLM=InternVL3-8B, Retriever=EVA-CLIP-8B, Feature=V+T2026.05 | 40.1 | |
| CgRAGLLM/MLLM=LLaVA-NeXT-7B, Retriever=EVA-CLIP-8B, Feature=V+T2026.05 | 39.9 | |
| CgRAGLLM/MLLM=LLaVA-NeXT-8B, Retriever=EVA-CLIP-8B, Feature=V+T2026.05 | 39.9 | |
| CgRAGLLM/MLLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B, Feature=V+T2026.05 | 39.8 | |
| MMKB-RAGRetriever=PreFLMR ViT-G, Model=Qwen2-VL-7B, Reranking=true2025.08 | 39.7 | |
| MMKB-RAGLLM/MLLM=Qwen2-VL-7B, Retriever=EVA-CLIP-8B, Feature=V2026.05 | 39.7 | |
| CgRAGLLM/MLLM=Qwen2-VL-7B, Retriever=EVA-CLIP-8B, Feature=V+T2026.05 | 39.6 | |
| KU-RAGLLM/MLLM=GPT-4o, Retriever=EVA-CLIP-8B, Feature=V+T2026.05 | 38.3 | |
| VLM-PRFLLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B2026.04 | 37.1 | |
| ReflectiVALLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B2026.04 | 36.8 | |
| EchoSightRetriever=EVA-CLIP-8B, Model=Mistral-7B, Reranking=true2025.08 | 35.5 | |
| ReflectiVARetriever=EVA-CLIP-8B, Model=LLaVA-MORE-8B2025.08 | 35.5 | |
| ReflectiVALLM/MLLM=LLaMA-3.1-8B, Retriever=EVA-CLIP-8B, Feature=V2026.05 | 35.5 | |
| DPRLLM=Multi-passage BERT, Retriever=CLIP ViT-B/322026.04 | 29.1 | |
| DPR_V+TLLM/MLLM=Multi-passage BERT, Retriever=CLIP ViT-B/32, Feature=V+T2026.05 | 29.1 | |
| VLM-PRFLLM/MLLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B, Feature=V+T2026.05 | 28.9 | |
| RET-2Retriever=RET-2, Model=LLaVA-MORE-8B2025.08 | 28.5 | |
| ReflectVALLM=LLaMA-3.1-8B, Retriever=EVA-CLIP-8B2026.04 | 28 | |
| GPT-4VLLM=GPT-4V, Retriever=–2026.04 | 26.9 | |
| GPT-4VMode=Zero-shot MLLMs2026.05 | 26.8 | |
| EchoSightLLM/MLLM=LLaMA-3.1-8B, Retriever=EVA-CLIP-8B, Feature=V2026.05 | 26.4 | |
| DBAgent (w/o SFT)LLM=Qwen2.5-VL-7B, Retriever=EVA-CLIP-8B + bge-m32026.04 | 23.6 | |
| LLaVA-NeXT-8BMode=Zero-shot MLLMs2026.05 | 22.5 | |
| LLaVA-NeXT-7BMode=Zero-shot MLLMs2026.05 | 22.1 | |
| Wiki-LLaVARetriever=CLIP-ViT-L, Model=LLaVA-1.5-7B2025.08 | 21.8 | |
| Qwen2.5-VL-7B (Base)LLM=Qwen2.5-VL-7B (Base), Retriever=–2026.04 | 21.7 | |
| Qwen2.5-VL-7BMode=Zero-shot MLLMs2026.05 | 21.7 | |
| EchoSightLLM=Mistral-7B/LLaMA-3-8B, Retriever=EVA-CLIP-8B2026.04 | 19.4 | |
| EchoSightLLM/MLLM=Mistral-7B/LLaMA-3-8B, Retriever=EVA-CLIP-8B, Feature=V2026.05 | 19.4 | |
| Wiki-LLaVALLM/MLLM=LLaMA-3.1-8B, Retriever=CLIP ViT-L/14 + Contriever, Feature=T2026.05 | 18.3 | |
| Qwen2.5-VL-3BMode=Zero-shot MLLMs2026.05 | 17.9 | |
| Wiki-LLaVALLM=LLaVA-v1.5-7B, Retriever=CLIP ViT-L/14 + Contriever2026.04 | 17.7 | |
| Wiki-LLaVALLM/MLLM=Vicuna-7B, Retriever=CLIP ViT-L/14 + Contriever, Feature=T2026.05 | 17.7 | |
| LLaVA-v1.5LLM=Vicuna-7B, Retriever=–2026.04 | 16.3 | |
| LLaVA-1.5LLM/MLLM=Vicuna-7B, Mode=Zero-shot MLLMs2026.05 | 16.3 | |
| Qwen2-VL-7BMode=Zero-shot MLLMs2026.05 | 16.2 | |
| LLaVA-1.5LLM/MLLM=LLaMA-3.1-8B, Mode=Zero-shot MLLMs2026.05 | 16 | |
| BLIP-2LLM=Flan-T5XL, Retriever=–2026.04 | 12.6 | |
| BLIP-2LLM/MLLM=Flan-T5XL, Mode=Zero-shot MLLMs2026.05 | 12.6 | |
| InstructBLIPLLM=Flan-T5XL, Retriever=–2026.04 | 11.9 | |
| InstructBLIPLLM/MLLM=Flan-T5XL, Mode=Zero-shot MLLMs2026.05 | 11.9 |