tinaface，eval result doesn't have esay-medium-hard result?

[wang-kangkang]

跑tinaface的评测，只有voc的ap，没有easy，medium，hard的结果： +-------+-------+----------+--------+-------+ | class | gts | dets | recall | ap | +-------+-------+----------+--------+-------+ | face | 31957 | 65363592 | 0.995 | 0.916 | +-------+-------+----------+--------+-------+ | mAP | | | | 0.916 | +-------+-------+----------+--------+-------+ 我想要easy medium hard的结果该怎么弄？

[hxcai]

@wang-kangkang please see the link to get easy, medium, hard reuslts.

[wang-kangkang]

@wang-kangkang please see the link to get easy, medium, hard reuslts.

我就是照着这个做的，才得到上面没easy-medium-hard的这种结果，能再说详细些吗？你是说跑那个matlab代码吗？ i was did by your link, and than get my result. can you be more specific? you mean the matlab code in eval_tools?

[hxcai]

@wang-kangkang Yes, you need to run the official matlab code in eval_tools to get results.

[wang-kangkang]

@wang-kangkang Yes, you need to run the official matlab code in eval_tools to get results.

thank you. i have another question. the result in wider_val hard is 0.934 or 0.924 in picture, but the result in wider_val hard is 0.930 or 0.922 in table. why are they different?

[hxcai]

@wang-kangkang The results in table are without TTA and the results in picture are our final score with TTA.

[wang-kangkang]

@wang-kangkang The results in table are without TTA and the results in picture are our final score with TTA.

thank you I have the third question. in test_widerface.py, i find the third param in build_dataloader() need to be 0, if use 1 by default, it will report segment fault. do you know why? the time of predict all Wider_val need almost 2 hours

[hxcai]

@wang-kangkang We haven't meet the error.