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
32 stars 6 forks source link

label site #17

Open assia855 opened 4 days ago

assia855 commented 4 days ago

Hello, thanks for sharing this project so interesting. I would like to create my own dataset, on the tutorial you didn't mention the website to label the 6D pose. thanks in advance.

cviviers commented 4 days ago

Hi @assia855,

Unfortunately there is no website. To make your own dataset is quite tricky. You need to:

  1. print the charuco board
  2. take pictures of it to get the camera calibration parameters
  3. place your object at a know offset from the charuco board coordinate system
  4. take pictures of it to project the 3D model of your object onto the scene
  5. obtain the 2d coordinate of your objects 3d bounding box

Alternatively you can do it all in simulation as discussed in #9.

assia855 commented 3 days ago

Hi @cviviers thanks for your answer. From what I understand I can use data_curation to create the label of my dataset I just need to create first the 3D model, to do that I should generate it with blender is that it? Thanks so much for your help.

cviviers commented 3 days ago

yes, that is a good start. I would say create a simple object first (simple cube) and make sure you understand the full process before trying intricate objects. Alternatively, if you have an RGB-D camera, you can follow this tutorial to skip making 3D objects yourself - https://github.com/F2Wang/ObjectDatasetTools

assia855 commented 1 day ago

thanks a lot @cviviers I will test that and update you with the results