前言
针对机器人抓取中的检测、分割、姿态识别、抓取点检测、路径规划等任务,总结了对应的数据集,在这里分享下,数据格式为类别+数量。
一、检测任务
PASCAL VOC:20类,11540
SUN:908类,131,072
MS COCO:91类,328,000
Places:434类,10 millions
Open Images:6000类,9 millions
二、分割任务
PASCAL VOC 2012 Segmentation:21类,2913
PASCAL-Context:540类,19,740
PASCAL-Part:20类,19,740
SBD:21类,11,355
MS COCO:80类,204,721
DAVIS:4类,8422
三、姿态识别任务
LineMod:15类,1100+ frame video sequences
T-LESS:30类,49K images
PU-APC:24类,10000 images
YCB-Video:21类,92 RGB-D videos
四、抓取点检测
Standford Grasping:10 object,13747 RGB Images,13747 Depth Images
Cornell Grasping:240 object,885 RGB Images,885 Depth Images
YCB Benchmarks:77 object,46200 RGB Images,46200 Depth Images
CMU dataset:150+object,50567 RGB Images
Google dataset:800000 RGB Images
Dex-Net 1.0:150+object,50567 RGB Images
Dex-Net 2.0:150+object,50567 RGB Images
JACQUARD:11619object,54485 RGB Images,108970Depth Images
五、抓取路径规划
抓取路径规划数据集:
1、Supersizingself-supervision: Learning to grasp from 50k tries and 700 robot hours.
2、Learning hand-eyecoordination for robotic grasping with deep learning and large-scale datacollection.
3、Multimodal grasp dataset: A novel visual–tactile data set for robotic manipulation.
抓取仿真:
1、Graspit! a versatile simulator for robotic grasping.
2、Opengrasp: A toolkit for robot grasping simulation.
3、Deep reinforcement learning for vision-based robotic grasping: Asimulated comparative evaluation of offpolicy methods.
上述内容,如有侵犯版权,请联系作者,会自行删文。