ResearchBenchmarksSingle-Phrase Grounding on EndoVis 18Follow76.67AccuracyVLM+SAM371.792473.058774.32575.5913Jun 14, 2026Evaluation ResultsMethodMethodLinksAccuracyVLM+SAM3Segmentation Module=SAM3Segmentation Module=SAM32026.0676.67VLM+SAMSegmentation Module=SAMSegmentation Module=SAM2026.0675.12VLM+SAM2Segmentation Module=SAM2Segmentation Module=SAM22026.0673.85VLM onlyTuning=LoRATuning=LoRA2026.0671.98