search

melonwan / denseReg

3D hand pose estimation via dense regression
179 stars 34 forks source link

Why there is nyu_bbx.pkl? #9

Open MeteorsHub opened 6 years ago

MeteorsHub commented 6 years ago

I found you use nyu_bbx when testing on NYU dataset, which means you use 'ground truth' hand location data when running test. I think it should be predict by network itself instead of given by you.

So how did you get nyu_bbx.pkl in your experiment? Is it computed from ground truth xyz labels or predicted by another method which can compute bbx of hand location?

melonwan commented 6 years ago

Hi, the bounding box is computed by a hand detector, which is based on a 2D hour-glass estimation.

MeteorsHub commented 6 years ago

Is the hand detector in this release?

As far as i understand, you mean first train a hourglass net to predict 2d cordinate of all joint points, and then compute a bbox to surround all joint points?

Is this part mentioned on your paper and considered in the running speed?

melonwan commented 6 years ago

this part is not included but is very simple to implement. You could also use the deep prior++ method which gives more accurate result according to my previous experiment. The time for this part is not included into the running time calculation.

MeteorsHub commented 6 years ago

Ok, thank you for the reply

melonwan commented 6 years ago

thank you too for the valuble suggestions. I'll later clean the code again. Sorry for the inconvenience brought by.