Motion Human-Likeness Assessment on Motion Turing (test)
0.5813MAEPTR-Net
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PTR-Net2026.03 | 0.5813 | 0.7926 | 0.6841 | |
| MotionBERTEvaluation Protocol=Fine-tuned2026.03 | 0.6252 | 0.8465 | 0.6142 | |
| TransformerModel Configuration=Lightweight2026.03 | 0.6387 | 0.8259 | 0.5728 | |
| MotionBERTEvaluation Protocol=Frozen Backbone2026.03 | 0.6846 | 0.9025 | 0.5315 | |
| Gemini 2.5 ProPrompting Strategy=Prototype-Driven Evaluation (PDE), VLM-based evaluation without task-specific training=true2026.03 | 1.2616 | 1.5397 | 0.2188 | |
| Gemini 2.5 ProPrompting Strategy=Posture-Aware Chain-of-Thought (PA-CoT), VLM-based evaluation without task-specific training=true2026.03 | 1.2682 | 1.5214 | 0.2303 | |
| Gemini 2.5 ProPrompting Strategy=Direct Evaluation (DE), VLM-based evaluation without task-specific training=true2026.03 | 1.3105 | 1.5873 | 0.1609 | |
| Gemini 2.5 ProPrompting Strategy=Context-Guided Evaluation (CGE), VLM-based evaluation without task-specific training=true2026.03 | 1.3314 | 1.5986 | 0.1658 | |
| Gemini 2.5 ProPrompting Strategy=Direct Evaluation Chain-of-Thought (DE-CoT), VLM-based evaluation without task-specific training=true2026.03 | 1.604 | 1.9146 | 0.1353 | |
| Qwen3-VL-PlusModel Configuration=shot-based, VLM-based evaluation without task-specific training=true2026.03 | 1.7714 | 2.1018 | — |