centerpose icon indicating copy to clipboard operation
centerpose copied to clipboard

PE accuracy

Open Dantju opened this issue 4 years ago • 16 comments

@tensorboy 你好,看你的表格里面DLA-34在coco val上的精度是62.7%,我这边用官方的model测出来是59.2%,请问你做了什么别的优化措施嘛

Dantju avatar Mar 09 '20 02:03 Dantju

@tensorboy 请问为什么evalute中的载入的val数量只有2693,coco val中总共有5000才对

Dantju avatar Mar 09 '20 05:03 Dantju

明天回答你可以吗,因为只有2693张里面有人。

On Sun, Mar 8, 2020 at 10:45 PM Dantju [email protected] wrote:

@tensorboy https://github.com/tensorboy 请问为什么evalute中的载入的val数量只有2693,coco val中总共有5000才对

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/tensorboy/centerpose/issues/27?email_source=notifications&email_token=AEDERP2W7ARXZ36DHIP5IDDRGR7BHA5CNFSM4LD7PPE2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEOFXJUI#issuecomment-596341969, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEDERPYI36WOMVJR33HBMNDRGR7BHANCNFSM4LD7PPEQ .

tensorboy avatar Mar 09 '20 05:03 tensorboy

@tensorboy ok 对了你的centerpose相对于centernet都做了哪些改进了 除了换了几个backbone之外

Dantju avatar Mar 09 '20 07:03 Dantju

@Dantju 你好,我其实并没有太多的改进,就是调了调参数,DLA34对应的参数都在这里:https://github.com/tensorboy/centerpose/blob/master/experiments/dla_34_512x512.yaml

tensorboy avatar Mar 10 '20 03:03 tensorboy

@tensorboy 调了参数之后AP有上升吗

Dantju avatar Mar 10 '20 05:03 Dantju

当然有了,只要不断调就会越来越高。

On Mon, Mar 9, 2020 at 10:42 PM Dantju [email protected] wrote:

@tensorboy https://github.com/tensorboy 调了参数之后AP有上升吗

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/tensorboy/centerpose/issues/27?email_source=notifications&email_token=AEDERPYVCYR32FARLKSD66LRGXHK5A5CNFSM4LD7PPE2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEOKC5HY#issuecomment-596913823, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEDERP2E5Q44OF7URGOXR7LRGXHK5ANCNFSM4LD7PPEQ .

tensorboy avatar Mar 10 '20 06:03 tensorboy

@tensorboy 主要调了哪些参数呢

Dantju avatar Mar 10 '20 09:03 Dantju

如果都枚举出来的话,其实我也忘的差不多了。。。反正能看到不少提高的地方。。。

tensorboy avatar Mar 10 '20 21:03 tensorboy

@tensorboy 请问一下centernet是有两个分支,一个PD,一个PE,你有没有发现同时训练PD_PE中的PD的效果是差于单独训练的PD的效果,你有研究如何改善这个问题吗,如果能改善PD_PE中的PD是否会对PE也有促进效果呢

Dantju avatar Mar 12 '20 02:03 Dantju

PE和PD是什么的缩写呢。@Dantju

tensorboy avatar Mar 12 '20 02:03 tensorboy

person detection 和 Pose Estimation

Dantju avatar Mar 12 '20 03:03 Dantju

@tensorboy

Dantju avatar Mar 12 '20 03:03 Dantju

你的centerpose不就是做PD和PE 你有用centetnet单独训练过PD吗

Dantju avatar Mar 12 '20 03:03 Dantju

@tensorboy 请问主要调了哪些参数呢,哪些是主要的参数

Dantju avatar Mar 25 '20 02:03 Dantju

@sulthanallaudeen this is scam. don't interact

salemalem avatar Jan 01 '24 01:01 salemalem

@sulthanallaudeen this is scam. don't interact

Block and report this scammer account.

Gh0stExp10it avatar Jan 01 '24 01:01 Gh0stExp10it