search

YuliangXiu / ECON

[CVPR'23, Highlight] ECON: Explicit Clothed humans Optimized via Normal integration
https://xiuyuliang.cn/econ
Other
1.06k stars 105 forks source link

Questions about testing #105

Closed glorioushonor closed 8 months ago

glorioushonor commented 8 months ago

Thank you for your excellent work, but I noticed an error occurring during the testing phase. Could it be that the Cape dataset is corrupted? I've tried re-downloading the Cape dataset, but it doesn't seem to resolve the issue. image

YuliangXiu commented 8 months ago

You could locate the problematic subject by try-catch. The evaluation works well on my end on the same data.

glorioushonor commented 7 months ago

You could locate the problematic subject by try-catch. The evaluation works well on my end on the same data.

This error is quite strange; when I train again, sometimes this error does not occur, but the testing results show a discrepancy of 0.1cm compared to the data in the paper, which is a significant difference. The cause of this error is also attributed to the use of pytorch3d==0.7.1. When using pytorch3d==0.7.5, a segmentation fault occurs. Therefore, pytorch==0.7.2 is recommended for use. In the end, testing with the pre-trained model yields results with a maximum difference of only 0.003 cm.