Hello everyone,
I am experimenting with my new Hailo 8L device on Raspberry PI 5. I want to use Person Re-Id for one of my projects. I have downloaded a precompiled model repvgg_a0_person_reid_512.hef which is compiled for Hailo 8L device. When I try to use it using GStreamer, the post-processing looks for the output layer repvgg_a0_person_reid_2048.
Can anyone guide me on how to find post-processing so for repvgg_a0_person_reid_512 model.
Hey @ap20feb,
Welcome to the Hailo Community!
You can find all the postprocessing resources here. Please note that you might need to adjust them to fit your specific use case: Hailo Model Zoo - Postprocessing
For your specific model, refer to this link: Person Re-ID Postprocessing
My application is built on TAPPAS and hence I would require post-processing .SO file.
The above link is for python post-processing which is not yet available on Raspberry Pi.
Yeah, my bad.
You can find all the available .so files after you install Tappas at the following location:
apps/h8/gstreamer/libs/post_processes/
Unfortunately, we couldn’t create .so files for every model. If you can’t find the one you’re looking for, you can create one yourself. Please refer to the guide: Write Your Own Post-Process
I’m not sure if you’re currently experiencing the same problem as I’ve had, but to use repvgg_a0_person_reid_512, you have to edit the post-processing file in tappas by cloning the repo from github, change the "#define OUTPUT_LAYER_NAME "repvgg_a0_person_reid_2048/fc1"by #define OUTPUT_LAYER_NAME "repvgg_a0_person_reid_512/fc1"in core>hailo>libs>postprocesses>recognition>repvgg.cpp and then recompile the file.
Hi @axplessis. Thanks to your post, I was able to implement it successfully. Now it is working. Thanks again 