An error while converting file from onnx to har or hef

General
,
1

Hey everyone, So I’ve been trying to convert this onnx file to hef or har, but whenever I try doing it, I keep getting an error saying that the hailo packages are not there. And when I check all the packages, things like the hailort, compiler, etc are installed, but this one command hailo_convert doesn’t exist, I checked chatgpt and other online forums, but couldn’t resolve this issue, would really appreiciate it if any of u could help with this, thanks!!

File and file paths:

  • python source code, compiler (whl file), hailort (whl file), ai software suite (run file) :- All r in: Home/Downloads
  • hailort (deb file), hailo_venv (Virtual environment), the tflite and onnx file obtained from the python code:- All r in Home/Downloads/software_ai_sw_suite
  • COMPRESSED (File containing test, train and valid folders with the xml and jpg image data): All in Home/Trial
2

Hey @Shannen_Milton ,

Welcome to the Hailo Community!

I noticed there might be some confusion with the command you’re using. For DFC operations, you should use the hailo command rather than hailo_convert in the CLI. If you’re trying to run this from a Python script, could you share it so I can provide more specific guidance? Also, please ensure you’re working within the virtual environment where all the necessary packages are installed.

3

Thanks for replying back, and idk why but the compiler keeps giving me the same error even now, And yes, Im using a python script (I managed to train the data, but yea…not able to convert it to hef). Will add it below

File path:
/home/ubuntu/optpascaltotflite_part2.py

import os
import datetime

from tflite_model_maker.config import ExportFormat
from tflite_model_maker import model_spec
from tflite_model_maker import object_detector

import tensorflow as tf
assert tf.__version__.startswith('2')

tf.get_logger().setLevel('ERROR')
from absl import logging
logging.set_verbosity(logging.ERROR)

spec = model_spec.get('efficientdet_lite0')

# Specify your data directories (updated paths)
train_image_dir = '/home/ubuntu/Downloads/COMPRESSED/train'
train_annotations_dir = '/home/ubuntu/Downloads/COMPRESSED/train'
validation_image_dir = '/home/ubuntu/Downloads/COMPRESSED/valid'
validation_annotations_dir = '/home/ubuntu/Downloads/COMPRESSED/valid'
test_image_dir = '/home/ubuntu/Downloads/COMPRESSED/test'
test_annotations_dir = '/home/ubuntu/Downloads/COMPRESSED/test'

# Load data using from_pascal_voc method
train_data = object_detector.DataLoader.from_pascal_voc(
    train_image_dir, train_annotations_dir, label_map={1: "person"}
)
validation_data = object_detector.DataLoader.from_pascal_voc(
    validation_image_dir, validation_annotations_dir, label_map={1: "person"}
)
test_data = object_detector.DataLoader.from_pascal_voc(
    test_image_dir, test_annotations_dir, label_map={1: "person"}
)

# Create and train the object detection model
print("**********************Training Model*********************************")
print("Timestamp", datetime.datetime.now())

model = object_detector.create(
    train_data,
    model_spec=spec,
    epochs=150,
    batch_size=10,
    train_whole_model=True,
    validation_data=validation_data
)

# Evaluate the model
print("**********************Evaluating Model*********************************")
print("Timestamp", datetime.datetime.now())

model.evaluate(validation_data)

# Export the trained model to TFLite format
print("**********************Exporting to TFLite*********************************")
print("Timestamp", datetime.datetime.now())

modelname = 'detect_Rasp5_fishery_906img.tflite'
model.export(export_dir='.', tflite_filename=modelname)

# Evaluate the exported TFLite model on the test set
print("**********************Evaluating TFLite Model on Test Set*********************************")
print("Timestamp", datetime.datetime.now())

model.evaluate_tflite(modelname, test_data)