Text Retrieval on DynPose-100K 1.0 (test)
46.3Top-1 AccuracyCamFormer embeddings (ViPE)
Evaluation results
| Method | Links | |
|---|---|---|
| CamFormer embeddings (ViPE)Modality=trajectory, Pose Source=ViPE2025.11 | 46.3 | |
| CamFormer embeddings (original)Modality=trajectory, Pose Source=original2025.11 | 36.2 | |
| ShotVL-7BModality=cam desc. (text)2025.11 | 33.1 | |
| Qwen2.5-VL-7B on CameraBenchModality=cam desc. (text)2025.11 | 27.8 | |
| SFT Qwen3-4B-InstructModality=trajectory (as text)2025.11 | 25.7 | |
| Gemini-3-ProModality=trajectory (as video)2025.11 | 24.7 | |
| SFT Qwen3-VL-4BModality=trajectory (as video)2025.11 | 23.1 | |
| Qwen3-VL-32B-InstructModality=trajectory (as video)2025.11 | 21.5 |