Intent reasoning on Real-world intent reasoning 88 samples (test)
94.3AccuracyGesVLA (VLMint)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GesVLA (VLMint)Module=VLMint2026.05 | 94.3 | 97.2 | |
| Baseline-2 (geometric pipeline)Pipeline Components=MediaPipe, GroundingDINO, DepthAnythingV22026.05 | 59.1 | 78.4 | |
| Baseline-1 (prompted MLLM)Base Model=Qwen3.5-plus2026.05 | 38.6 | 61.4 |