Filter-out-class-ids param doesn't work for ds-tao-segmentation app

• Hardware Platform: JETSON ORIN
• DeepStream Version: 6.2
• JetPack Version: 5.1-b147
• TensorRT Version: 5.1
• Issue Type: bug

I have to use a custom app (provided by the NVidia team on GitHub) [ app: deepstream_tao_apps/configs/unet_tao at master · NVIDIA-AI-IOT/deepstream_tao_apps · GitHub] since I am porting over TAO models to deepstream. I am also trying to remove/ignore a certain class on the output, but the parameter does not seem to be working. In the APP github link i put above, i am only putting “filter-out-class-ids: 1” under the property key… could you direct me to what is going on or what i can even do to further investigate the issue?

It’s for detected model: Filter out detected objects belonging to specified class-ids. The unet_tao is a segmentation model which filtering is not supported currently.

It might be helpful to spit out error messages if that’s the case, I spent a little too long digging into the SDK just to find out here it’s not compatible. In addition, why are there all of these separate apps I have to manually find and run when it should just be all in one with “deepstream-app”. Especially if the software is going to be proprietary and not open source I would expect a much more consistent design.

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

We’ll consider to add some warning messages later. Currently, our guide provides detailed instructions about this field. filter-out-class-ids

We have some apps that show specific scenarios separately for customers to use conveniently.
https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_C_Sample_Apps.html

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