BerkeleyAutomation / gqcnn

Python module for GQ-CNN training and deployment with ROS integration.

Home Page:https://berkeleyautomation.github.io/gqcnn

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Issue: Bug/Performance Issue [Custom Images] - point cloud is not smooth

leeivan1007 opened this issue · comments

  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 16.04
  • Python version: 3.7.5
  • Installed using pip or ROS: pip
  • Camera: realsenseD415
  • GPU model (if applicable): Nvidia 1080 Ti

It is good by 'GQCNN-4.0-PJ' and always can grasp successfully.
But it can't predict successfully by 'GQCNN-4.0-SUCTION'.

Our test image

the depth distance is 0.693m with Q=0.185
image

image

image

there have a point cloud file:
https://drive.google.com/file/d/13iyM7MnbWEN_LE1hMO_S_Cwu3fr_XLZC/view?usp=sharing

The original image [ depth_0.npy ]

image

image

image

When grasp_candidates is opened

sampled grasps iter 0
image

sampled grasps iter 1
image

sampled grasps iter 2
image

Question

The original image's surface is smooth. I think maybe it is the cause that suction method can't be successful. Maybe I miss something, but now i don't know what happen clearly...

Think for any suggestions!