Run Yolov8 for Python 3.6

Question

Yolov8 is build to run with 3.8 or above and pip install ultralytics is not compatible with prior versions.

I have the constraint of using a prior version of Python to execute the code on a microcomputer like Jetson Nano or Jetson Xavier.

I've seen on many issues of Yolov8 repository people asking how to solve this issue aswell.

I'll answer bellow, the solution I have found so that others can benefit it or improve it.

Answer 1

First start by cloning the ultralytics repository git clone https://github.com/ultralytics/ultralytics.git into your lib folder.

Then do the following modifications:

1. In "[your_path]/lib/ultralytics/ultralytics/utils/__init__.py"

• Comment import importlib.metadata line 4 replace it by import pkg_resources
• Replace importlib.metadata.version("torchvision") by TORCHVISION_VERSION = pkg_resources.get_distribution("torchvision").version line 47

This is because importlib.metadata is not present in python 3.6

2. In "[your_path]/lib/ultralytics/ultralytics/utils/checks.py",

• Comment import importlib.metadata line 13

3. in “[your_path]/lib/ultralytics/ultralytics/hub/auth.py”

• replace if header := self.get_auth_header(): line 93 by

header = self.get_auth_header() if header:

This is because the walrus operator is not present in python 3.6

WARNING

This might break some things in the way yolov8 is working. But from what I have tested it is sufficient to run the predictions.