search

HuangJunJie2017 / UDP-Pose

Official code of The Devil is in the Details: Delving into Unbiased Data Processing for Human Pose Estimation
Apache License 2.0
307 stars 54 forks source link

Confuse about merge origin hrnet into mmpose #31

Closed Huuush closed 3 years ago

Huuush commented 3 years ago

Sry to bother. I use UDP for cloth landmark detection, and I implement it in your origin repo. But when I try to merge it in to mmpose, the mAP is not nomal when compare with origin implementation. Do you have any advice ? I'm new to mmpose.( I implement my dataset by inheriting the 'Topdowndataset' instead of origin Jointdataset.) Two more questions:
1.Btw, I notice that the implementation of the hyperparameter: KPD is different between origin and mmpose. In mmpose, we must calculate it by: valid_radius = factor * heatmap_size[1], so does it mean that i must change the factor if I got a different heatmap_size?
2.I wonder if I need to change the KPD for my own dataset? KPD=3.5 is just suitable for human pose? Thanks a lot !!! :)

HuangJunJie2017 commented 3 years ago

'the mAP is not normal' to what extent?

HuangJunJie2017 commented 3 years ago

as far as i know , kpd should be adjusted when the heatmap size is changed (a linear dependence).

Huuush commented 3 years ago

the origin is 0.688 and the mmpose implementation is 0.652

HuangJunJie2017 commented 3 years ago

you can try heatmap only supervision in mmpose, both combined target and gaussian heatmap only target are provided in mmpose

Huuush commented 3 years ago

as far as i know , kpd should be adjusted when the heatmap size is changed (a linear dependence).

I notice that the kpd is set to 5.25 for 384x288. How about 384x512? (嗐，谁让图给得越大 效果越好呢）

HuangJunJie2017 commented 3 years ago

emmm, the difference is not that large. You can convert the trained models into mmpose format, (adjust the keyname of parameters) to check the inference at first.

HuangJunJie2017 commented 3 years ago

as far as i know , kpd should be adjusted when the heatmap size is changed (a linear dependence).

I notice that the kpd is set to 5.25 for 384x288. How about 384x512? (嗐，谁让图给得越大 效果越好呢）

kpd is set to 3.5 for 384x512 cloth landmark detection in original hrnet repo?

Huuush commented 3 years ago

as far as i know , kpd should be adjusted when the heatmap size is changed (a linear dependence).

I notice that the kpd is set to 5.25 for 384x288. How about 384x512? (嗐，谁让图给得越大 效果越好呢）

kpd is set to 3.5 for 384x512 cloth landmark detection in original hrnet repo?

When the first time I introduce UDP to my own cloth landmark detection task, I set kpd to 3.5 acrroding to the readme, and now I know how to deal with it. Thanks for your wonderful work ! And thanks for your reply!!!

Huuush commented 3 years ago

Did you mind to tell me the result about UDP in RSN? Curious about it.