Merge Vision Recognition Database

Posted over 2 years ago by George Apergis

Post a topic
George Apergis

Is there a way to merge two vision recognition databases? More specifically, the .vrd files.

0 Votes


Sorted by

George Apergis posted over 2 years ago

After some .png and .yaml renaming and  .yaml editing, the merging was successful. 
Thanks for the tips.

0 Votes


Clement Portales posted over 2 years ago Admin

Hello George, 

So I've asked around and this isn't something that has been done before in SBRE and the experts/creators of this format have now left the company. However, someone over at SBRE was kind enough to give some insights on how it could be done, by having a look at the code of the .vrd file. Be aware that the method indicated isn't guaranteed to work ! It is merely a theory formulated after the code was looked at. 

Here's a retranscription of the info I could gather : 

For each objects database, this format is made of a file with all the required information : 

  • A file .yaml containing meta-data and objects models (named ${UUID}.yaml)
  • Images ${UUID}.${n}.png which define the objects models

This file is compressed into a tar.gz format by Choregraphe, the resulting file gets the extension .vrd, therefore a .vrd is a .tar.gz.

Schematically, in order to merge 2 .vrd files, you'd need to : 

  • unzip the .vrd files with tar -xzf ${files}.vrd
  • Merge together the .yaml files and the images within the different files and coordinating the index between .yaml files and images
  • Reconstitute the database by zipping again the merged files into a .tar.gz, rename it a .vrd with tar -czf.

Hope this could give you a few leads on how to solve your issue ! 

0 Votes

Login or Sign up to post a comment