search

kmittle / Grasp-Detection-NBMOD

This is the repository for the NBMOD dataset and the code of the paper "NBMOD: Find It and Grasp It in Noisy Background.
MIT License
37 stars 2 forks source link

Inference on GraspNet1Billion dataset #1

Closed elenacliu closed 10 months ago

elenacliu commented 10 months ago

Hello author,

I tried your test code on GraspNet1Billion dataset, but got the results. It may be wrong?

image
elenacliu commented 10 months ago

I find the reason. I should use the (300,400,3) size image to draw the box on.

Andyyoung0507 commented 9 months ago

I find the reason. I should use the (300,400,3) size image to draw the box on. Why the image size should be (300,400,3), I use my own picture to generate the grasp, the results are also strange. Only one box but beyond the desired object. @elenacliu

kmittle commented 9 months ago

The size I used in my experiment was 416 x 416. You can modify the meshing method to detect images of any size.

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2024年1月5日(星期五) 中午11:42 收件人: @.>; 抄送: @.***>; 主题: Re: [kmittle/Grasp-Detection-NBMOD] Inference on GraspNet1Billion dataset (Issue #1)

I find the reason. I should use the (300,400,3) size image to draw the box on. Why the image size should be (300,400,3), I use my own picture to generate the grasp, the results are also strange. Only one box but beyond the desired object.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

Andyyoung0507 commented 9 months ago

The size I used in my experiment was 416 x 416. You can modify the meshing method to detect images of any size. … ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2024年1月5日(星期五) 中午11:42 收件人: @.>; 抄送: @.>; 主题: Re: [kmittle/Grasp-Detection-NBMOD] Inference on GraspNet1Billion dataset (Issue #1) I find the reason. I should use the (300,400,3) size image to draw the box on. Why the image size should be (300,400,3), I use my own picture to generate the grasp, the results are also strange. Only one box but beyond the desired object. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.>

Thank you. I am sorry about that I didn't find the meshing method in the code repository and the paper, I just modified the boxes value by scaling up the ratio between the input image size and 416, but the results seems not so well. What should I do to improve the performance of generated grasp? Thanks!

1