search

takumayagi / fpl

Code/data of the paper "Future Person Localization in First-Person Videos" (CVPR2018)
https://artilects.net/projects.html
MIT License
66 stars 16 forks source link

prepare dataset #4

Open bbidong opened 5 years ago

bbidong commented 5 years ago

Hi @takumayagi , Sometimes the output of open-pose less than 18 joint points, but the pose stream of the input in your network must be 18*2, so how do you deal with the rest joint points???? Looking forward to your reply!

takumayagi commented 5 years ago

Hi @bbidong , We imputed the missing values from other detection using nearest neighbor imputation (we used fancyimpute in our implementation). Please refer to Section 4.2 of my thesis for further details. https://artilects.net/theses/tyagi_master_thesis.pdf

Guanwh commented 5 years ago

Hi @bbidong ,

Have you mastered how to extract features? If so, can you give me some advice?

Looking forward to your reply!

bbidong commented 5 years ago

@Guanwh yeap, i am happy to share with you

trungmanhhuynh commented 4 years ago

Hi @bbidong, can you also share with me how you impute the missing key points? Do the missing keypoints degrade the accuracy on your custom datasets? Thanks

Namechaoo commented 1 year ago

@bbidong Hi, I'm having some difficulties with location and scale extraction, can you help me, please?