RobotLocomotion / pytorch-dense-correspondence

Code for "Dense Object Nets: Learning Dense Visual Object Descriptors By and For Robotic Manipulation"

Home Page:https://arxiv.org/pdf/1806.08756.pdf

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Misleading depth scaling factor "DEPTH_IM_SCALE"

Zador-Pataki opened this issue · comments

Hi, I am referring to lines 499:
depth_vec = torch.index_select(img_a_depth_torch, 0, uv_a_vec_flattened)*1.0/DEPTH_IM_SCALE
and 595:
depth2_vec = torch.index_select(img_b_depth_torch, 0, uv_b_vec_flattened)*1.0/1000
in file correspondence_finder.py

If you change the depth scale variable DEPTH_IM_SCALE, there will be an incorrect filtering of correspondences