Point Grounding on InstructPart (test)
90Patch AccuracyMolmo-7B Ours
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Molmo-7B OursModel Category=MLLM w/ pointing ability, Pointing Method=Ours2026.06 | 90 | 86.8 | |
| Qwen2.5-VL-7B OursModel Category=MLLM w/ pointing ability, Pointing Method=Ours2026.06 | 87.7 | 81.8 | |
| First-Gen-MLLM OursModel Category=MLLM w/o pointing ability, Pointing Method=Ours2026.06 | 80.3 | 78.3 | |
| Molmo-7B text pointingModel Category=MLLM w/ pointing ability, Pointing Method=text pointing2026.06 | 73.7 | 71 | |
| Qwen2.5-VL-7B text pointingModel Category=MLLM w/ pointing ability, Pointing Method=text pointing2026.06 | 72.2 | 70.8 | |
| LISA-7BModel Category=Segmentation-based Models2026.06 | 48 | 46.4 | |
| Molmo-7B attention pointingModel Category=MLLM w/ pointing ability, Pointing Method=attention pointing2026.06 | 46.8 | 37.8 | |
| GLaMM-7BModel Category=Segmentation-based Models2026.06 | 41.2 | 40 | |
| Qwen2.5-VL-7B attention pointingModel Category=MLLM w/ pointing ability, Pointing Method=attention pointing2026.06 | 35.2 | 28.3 | |
| First-Gen-MLLM attention pointingModel Category=MLLM w/o pointing ability, Pointing Method=attention pointing2026.06 | 22.7 | 19.4 | |
| X-DecoderModel Category=Segmentation-based Models2026.06 | 18.5 | 17.8 | |
| First-Gen-MLLM text pointingModel Category=MLLM w/o pointing ability, Pointing Method=text pointing2026.06 | 4 | 3.3 | |
| VLPartModel Category=Segmentation-based Models2026.06 | 0.8 | 0.8 |