How to use Instance Segmentation without GStreamer

Hello!

I was looking through the example pipeline for instance segmentation (specifically with yolov5 segmentation) and wondering how I can use the pipeline without GStreamer and instead my own pre processing method. To my understanding it looks like these pipelines are interwoven with pre processing and post processing steps using GStreamer. The end goal is to pre-process a single image without GStreamer and output data from the model (no post processing). How can I reduce the pipeline to still do inference without GStreamer? Thanks!

Hey @Zaba ,

Welcome to the Hailo Community!

For alternatives to GStreamer, check out our examples repository:

The repository includes both Python and C++ implementation examples for instance segmentation that should help you get started. Feel free to explore them and let me know if you need any assistance!

Thanks! Was trying to work on this example and it says I need to import the model_zoo. The issue with this is I am attempting to run this off of the pi-5, and the model zoo requires much more powerful hardware when looking at the documentation, as well as on the software downloads page. Is there a way around this, and do I need to download the entire model zoo library to run the instance segmentation example?

I managed to get it working, just had to wipe and reinstall the drivers!

1 Like

Hey @Zaba, I’m having similar problems running the python implementation of instance segmentation on RPi5.

I can’t install hailo_model_zoo withouth the hailo_dataflow_compiler which isn’t available for RPi platform.

Could you please explain in a little more detail how did you managed to get it running please?

Best regards,
Radek

Hi @Zaba , Have you compared the accuracy between running with GStreamer and running with OpenCV?

I switched to using this example code from GitHub, but its accuracy is very poor. However, when I use GStreamer, everything works perfectly fine. Have you encountered this issue before?