Object Counting on LVIS-Count 100-image zero-shot (test)
23.36MAEVTOS Analyzer ON
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VTOS Analyzer ONFamily=VTOS (ours), Analyzer=ON, LLM=Sonnet 4.6, VFM=Grounding DINO + SAM2026.06 | 23.36 | 37.77 | 4.02 | 40.48 | |
| VTOS Analyzer OFFFamily=VTOS (ours), Analyzer=OFF, LLM=Sonnet 4.6, VFM=Grounding DINO + SAM2026.06 | 32.83 | 47.86 | 16.39 | 36.75 | |
| Sonnet 4.6 (Count Only)Family=End-to-End VLM, Mode=Count Only2026.06 | 35.13 | 56.88 | 4.35 | — | |
| Sonnet 4.6 (Direct Bbox)Family=End-to-End VLM, Mode=Direct Bbox2026.06 | 35.18 | 42.31 | 33.64 | 7.54 | |
| CodeAct (Sonnet 4.6 + tools)Family=Agentic, Components=Sonnet 4.6 + tools2026.06 | 39.38 | 49.66 | 32.4 | 29.21 | |
| CountGD (vanilla)Family=Specialist, Variant=vanilla2026.06 | 43.01 | 126.82 | 32.05 | 36.13 | |
| Reflexion (Sonnet 4.6 + visual critic)Family=Agentic, Components=Sonnet 4.6 + visual critic2026.06 | 44.28 | 56.54 | 25.88 | 24.91 | |
| ReAct (Sonnet 4.6 + DINO)Family=Agentic, Components=Sonnet 4.6 + DINO2026.06 | 46.31 | 52.6 | 46.03 | 19.66 | |
| SAHI (DINO)Family=Heuristics, Base Model=DINO2026.06 | 46.57 | 50.42 | 44.65 | 15.91 | |
| Qwen3-VL-32B (Direct)Family=End-to-End VLM, Mode=Direct, Model Size=32B2026.06 | 48.32 | 47.26 | 47.26 | 9.98 | |
| CountGD (Crop+NMS)Family=Specialist, Variant=Crop+NMS2026.06 | 48.84 | 156.13 | 42.52 | 36.59 | |
| Grounding DINO (thr=0.10)Family=Specialist, Threshold=0.102026.06 | 52.76 | 77.61 | 27.44 | 32.19 | |
| Qwen3-VL-8B (Direct)Family=End-to-End VLM, Mode=Direct, Model Size=8B2026.06 | 54.26 | 54.26 | 54.26 | 3.38 |