Custom strcut for detection or an extension for NvDsInferObjectDetectionInfo

Please provide complete information as applicable to your setup.
• DeepStream Version
Deepstream 6.2

• Issue Type( questions, new requirements, bugs)
Query / New requirements

• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)
I have a detection problem which outputs the angle of the bounding box as well. The NvDsInferObjectDetectionInfo currently contains the following attributes: height, width, x, and. But I need a 5th attribute called angle. Or a way to give all the 4 (x, y) coordinates will suffice. Is there any way I can introduce a new struct to deepstream? Or is there any other way of modifying the existing struct without breaking deepstream?

We do not support it currently. Could you describe the specific purpose of your model? We’ll consider the way to implement it.

I have a model that outputs the bounding boxes along with the angle. center(x, y), height, width and an angle of the bounding box a. An object detection model for fish eye cameras. And additionally, we want to pass this information to a display sink as well for debugging. Current object metadata support is just the center, height and width. Which is insufficient to plot the exact bounding boxes or even use it for downstream applications. I’ve implemented the custom bounding box parser. But I don’t see a way to get the angle out of the bbox parser. Is there any other way to get this information out of the bbox parser?

Could you try to record this angle value yourself and attach it to the NvDsUserMetaList *obj_user_meta_list; of the NvDsObjectMeta structure? You can pass any structure with it.

1 Like

I’ve managed to attached the extra metadata to NvDsUserMeta following this and referring to deepstream/sources/apps/sample_apps/deepstream-infer-tensor-meta-test

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.