search

sfu-gruvi-3dv / deep_human

Code for iccv2019 paper "A Neural Network for Detailed Human Depth Estimation from a Single Image"
GNU General Public License v3.0
62 stars 8 forks source link

training data depth map #2

Closed xjturobocon closed 4 years ago

xjturobocon commented 4 years ago

Hey, Thanks for your sharing, good job! I found in the training dataset you provide, the folder 'old/3/', the depth info is in the format '.npy', the maxium depth is often 0.6, and the minimum number is negtive. the visualiztion of it shows it is very coarse in the part of torso. However, in the demo data, the maxium depth is bigger than 2, and all numbers aer positive. The visulization result is also finer than the train data.

Hope for your kindly explanation.

sta105 commented 4 years ago

The depth files in the folder 'old/3/' are actually pre-processed, we will re-upload our raw dataset. Thanks for reminding us.