search

cviviers / YOLOv5-6D-Pose

6-DoF Pose estimation based on the YOLOv5 framework. Specific focus on instruments in X-ray applications
https://ieeexplore.ieee.org/document/10478293
GNU General Public License v3.0
45 stars 6 forks source link

About dataset structure #8

Closed dionstew closed 3 months ago

dionstew commented 4 months ago

I am interested to this project and looking forward to implement the architecture using my dataset. I would like to know the dataset structure and the file it contains, if it is okay to share...

cviviers commented 4 months ago

Hi @dionstew,

The linemod dataset used in this work is a good example to start with. If you want to use it on your own data we provide some tools to get you going (I will improve these soon) . Additionally you can follow the code here.

Let me know if you have some specific question about the data curation.

dionstew commented 4 months ago

Hi @cviviers, thanks for the help.

Is 3d model data required? In my work, I try to estimate 6D poses on objects that do not necessarily have the exact same shape and size such as human body parts.

cviviers commented 4 months ago

For this line of work, yes, you do require some 3D model. Although you can likely change the model at test time if the object only varies in size (to be tested). However, I think what you are looking for might rather be something like human mesh recovery if you are planning on working with body parts.

cviviers commented 3 months ago

@dionstew can I close this issue? I assume the above mentioned line of work is more suitable for you.