Motion-to-Text on Motion-X
88.1RPrecision Top3Real
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Real2024.05 | 88.1 | — | — | — | — | — | — | — | — | |
| RealType=Ground Truth2024.05 | 88.1 | 68.1 | 82.4 | 2.897 | — | — | — | — | — | |
| M³GPT (Instruction-tuned)Training Strategy=Instruction-tuned, Hardware Platform=Ascend 910B NPU2024.05 | 84.6 | — | — | — | — | 11.52 | — | 42.64 | — | |
| M³GPTStrategy=Instruction-tuned2024.05 | 84.5 | — | — | — | — | 11.5 | — | 42.93 | — | |
| M³GPTTraining Phase=Fine-tuned2024.05 | 84.5 | 63.1 | 78.3 | 2.95 | 34.27 | 11.5 | 34.55 | 42.93 | 31.49 | |
| MotionGPT2024.05 | 84 | — | — | — | — | 11.21 | — | 31.36 | — | |
| M³GPTStrategy=Pre-trained2024.05 | 83.4 | — | — | — | — | 11 | — | 24.12 | — | |
| M³GPTTraining Phase=Pre-trained2024.05 | 83.4 | 62.7 | 77.3 | 2.946 | 33.31 | 11 | 34.1 | 24.12 | 30.96 | |
| M³GPT (Pre-trained)Training Strategy=Pre-trained, Hardware Platform=Ascend 910B NPU2024.05 | 83.1 | — | — | — | — | 11.05 | — | 24.03 | — | |
| TM2T2024.05 | 80.6 | — | — | — | — | 12.13 | — | 20.16 | — | |
| TM2T2024.05 | 80.6 | 57.4 | 72.6 | 2.988 | 30.54 | 12.13 | 32.52 | 20.16 | 25.37 | |
| Trained single taskStrategy=Single Task Training2024.05 | 76.7 | — | — | — | — | 10.14 | — | 22.92 | — | |
| Trained single task2024.05 | 76.7 | 56.5 | 70.6 | 3.011 | 31.07 | 10.14 | 31.65 | 22.92 | 28.19 | |
| Trained single taskTraining Strategy=single task training, Hardware Platform=Ascend 910B NPU2024.05 | 76.3 | — | — | — | — | 10.16 | — | 22.89 | — |