search

THU-VCLab / HGGD

Official code of paper "Efficient Heatmap-Guided 6-Dof Grasp Detection in Cluttered Scenes"
MIT License
51 stars 6 forks source link

question about the dataset #10

Closed Xiong5Heng closed 4 months ago

Xiong5Heng commented 4 months ago

Hello, Thanks for your work. I downloaded the dataset from the link https://cloud.tsinghua.edu.cn/d/e3edfc2c8b114513b7eb/, and there are 189 scenes in this zip file. However, I notice in your paper that you mention that training requires 500 scenes. My question is whether the final performance of the model trained using a data set of only 189 scenes can reach the value in the paper?

ChenThree commented 4 months ago

Thank you for your interst! We conduct 2 series of experiments in our paper. One is on Acronym dataset with your mentioned 500 scenes. Then the other one is on the GraspNet-1B dataset which is the 190 (0~189, 100 fo training and 90 for testing) scenes you downloaded.