Support for multiple objects

[ZJWdk]

Hello,

First, thank you for your brilliant work on this project! It's a very impressive tool.

I am trying to extend the functionality to estimate and visualize the orientation for multiple objects within a single image. I was particularly inspired by your "The Last Supper" demo, which handles this scenario perfectly.

My current workflow is as follows:

Use Grounded-Segment-Anything to generate masks for each target object.
Iterate through each mask, process it with Orient-Anything to predict its orientation.
Attempt to render the orientation axes for all objects and project them back onto the original image.
However, I'm facing challenges with the final rendering step. The output I'm able to generate does not correctly replicate the results from your demo(Below is my result). I suspect my issue lies in the rendering strategy—perhaps related to handling camera parameters across multiple objects or the projection logic itself.

I noticed the documentation left the "Best Practice" section blank. Would it be possible for you to share the script or provide some guidance on the correct method for rendering multiple objects, as was done for "The Last Supper" example?

Any help or pointers you could provide would be greatly appreciated. Thank you for your time and consideration!

[Yufang-Liu]

Did you fix the problem? I have the same question

[ZJWdk]

Not yet. I made changes with the help of Cursor, but could not achieve the performance of the demo. Maybe the new SAM3 can help