toteGrasping

Dataset for robotic grasping in dense cluster

Overview

This repository is the supplementary materials of the paper "GraspFusionNet: a two-stage multi-parameter grasp detection network based on RGB–XYZ fusion in dense clutter."Video by Liu et al. In this repo, we provide code of how to show grasp.

Results

Known objects

Novel objects

Datasets

Download links for our grasping dataset can be found on our toteGrasping.

Datasets Structure

• toteGrasping
  • toteGrasping/heightmap_color
    • toteGrasping/heightmap_color/%0.6d.png: color heightmap.
  • toteGrasping/heightmap_depth
    • toteGrasping/heightmap_depth/%0.6d_combine.npy: xyz heightmap, in meter.
  • toteGrasping/label_gPPN
    • toteGrasping/label_gPPN/%0.6d_bad|good.npy: good (x,y,1,angle,width,depth),angle:deg,width:mm,depth:meter/10000..
  • toteGrasping/label_gRPN
    • toteGrasping/label_gRPN/%0.6d.png: pixel wise label, graspable or non-graspable.
    • toteGrasping/label_gRPN/%0.6d.npy: pixel wise label, angle,width,depth.
  • toteGrasping/label_show
    • toteGrasping/label_show/%0.6d.png: original grasp label.
  • toteGrasping/train-split.txt: traing data
  • toteGrasping/test-split.txt: test data

Contact

If you have any questions, please let me know: sjtu-wenhai[at]sjtu[dot]edu[dot]cn