Report Generation on MIMIC (Factual Consistency and Completeness)
54.9Factual Consistency (FC)clean RAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| clean RAGModel=GPT-5, Retriever=clean RAG, Top-K=12026.05 | 54.9 | 56.86 | |
| LIARModel=GPT-5, Retriever=SigLIP, Top-K=12026.05 | 51.66 | 53.22 | |
| M3AttModel=GPT-5, Retriever=SigLIP, Top-K=12026.05 | 48.96 | 50.25 | |
| clean RAGModel=Gemini-2.5, Retriever=clean RAG, Top-K=12026.05 | 41.18 | 43.14 | |
| LIARModel=Gemini-2.5, Retriever=SigLIP, Top-K=12026.05 | 39.61 | 41.85 | |
| M3AttModel=Gemini-2.5, Retriever=SigLIP, Top-K=12026.05 | 38.12 | 40.63 | |
| w/o RAGModel=GPT-5, Retriever=w/o RAG, Top-K=12026.05 | 33.33 | 34.44 | |
| clean RAGModel=GPT-4o, Retriever=clean RAG, Top-K=12026.05 | 31.37 | 41.18 | |
| LIARModel=GPT-4o, Retriever=SigLIP, Top-K=12026.05 | 29.96 | 38.77 | |
| clean RAGModel=Claude-4.5, Retriever=clean RAG, Top-K=12026.05 | 29.41 | 39.22 | |
| M3AttModel=GPT-4o, Retriever=SigLIP, Top-K=12026.05 | 28.76 | 36.78 | |
| LIARModel=Claude-4.5, Retriever=SigLIP, Top-K=12026.05 | 26.46 | 35.49 | |
| M3AttModel=Claude-4.5, Retriever=SigLIP, Top-K=12026.05 | 24.31 | 32.7 | |
| clean RAGModel=LLaVA-Med, Retriever=clean RAG, Top-K=12026.05 | 19.61 | 15.69 | |
| w/o RAGModel=GPT-4o, Retriever=w/o RAG, Top-K=12026.05 | 18.89 | 19.44 | |
| LIARModel=LLaVA-Med, Retriever=SigLIP, Top-K=12026.05 | 18.36 | 14.81 | |
| M3AttModel=LLaVA-Med, Retriever=SigLIP, Top-K=12026.05 | 17.38 | 14.08 | |
| w/o RAGModel=Gemini-2.5, Retriever=w/o RAG, Top-K=12026.05 | 17.22 | 20.56 | |
| w/o RAGModel=Claude-4.5, Retriever=w/o RAG, Top-K=12026.05 | 12.22 | 12.78 | |
| w/o RAGModel=LLaVA-Med, Retriever=w/o RAG, Top-K=12026.05 | 7.22 | 4.45 |