Clarifications needed about using a custom dataset and yolov8 with Hailo

Hailo_model_zoo_v2.13.0.pdf: paragraph 10.4.4: it mentions I should use:
hailo_model_zoo/cfg/networks/yolov8s.yaml
and modify the input size in the preprocessing section of another file (referenced by the above):
hailo_model_zoo/cfg/base/yolo.yaml
BUT the preprocessing section says meta_arch: yolo_v5 instead of yolov8 ??
Using by onnx, should I change anything else, besides input size, on the above 2 files?
There is another file that is not specifcally mentioned:
./hailo_model_zoo/hailo_model_zoo/cfg/alls/generic/yolov8m.alls
that is problably used by something, that references another file
config/yolov8m_nms_config.json
where also the input size has to be changed

Does the the compile process add NMS to the pipeline? If not, how do I add it?

Can someone explain where exactly are all the yolov8 configuration files and which ones are used / for what purpose by the various tools?

Thanks

Hi @Thor,
Indeed there are multiple configuration and control files. Let me try to sort if out for you.
The yaml file is the Model-zoo configuration, it is the root that points to all other configuration and command files. See the below diagram.
The Model-Zoo further uses the Dataflow Compiler (DFC), the compile reads in an alls file as input commands, since NMS addition has many commands it was externalize into a seperate file.

Hi @Nadav, thanks.
I could not find any reference/info about:

yolo8x.yaml: what does the postprocessing: hpp: true mean/what are the options and where are they described

info: section. Is it just info or is actually used / what for/what should I change e.g. coco and other entries

Sure,

  • hpp - this entry tells if the model uses the hailort post-processing, or not.
  • info - this is just for information, stuff that we use in the model explorer for example.

We have this doc hailo_model_zoo/docs/YAML.rst at master · hailo-ai/hailo_model_zoo about the yaml, but it seems that we need to put some update on it. I’ll make sure we do.