Bug in the evaluation

[XiaoyuBIE1994]

Hi Droliven,

If I'm not wrong, there might be a small issue in the evaluation function. When iterating the data generator, if the number of multi-modal GT is 1, the evaluation will be bypassed #L280-L281, so in the final metric computation #L369-L37 the summed ADE/FDE ... should be divided by the real total number of data examples that have been used in the evaluation (should be a smaller number than i+1). After correcting this issue, the final results from the proposed pretrained models should be like this:

HumanEva
| ADE | FDE | MMADE | MMFDE |
| 0.234 | 0.247 | 0.350 | 0.327 |

Human3.6M
| ADE | FDE | MMADE | MMFDE |
| 0.378 | 0.495 | 0.483 | 0.525 |

Best,