I have a YoloV8 data file format that is an annotation of data (images) done manually.
What is the most effective and straightforward way of generating a model and therefore yielding the weights file? is it using darknet through the command:
darknet.exe detector train data/obj.data yolo-obj.cfg backup\yolo-obj_2000.weights
then something like the following to generate the associate model:
python tools/model_converter/convert.py cfg/yolov3.cfg weights/yolov3.weights weights/yolov3.h5
Or using Roboflow through:
version.deploy(model_type="yolov8", model_path=f”{HOME}/runs/detect/train/”)
Seems to me darknet is more difficult to install.
>Solution :
Seems to me darknet is more difficult to install.
Now why would you say something like that?
The build and installation steps are very well documented. Make sure you are using the most recent version of Darknet/YOLO: https://github.com/hank-ai/darknet#table-of-contents
If you find the steps difficult or lacking in any way, please let me know on the Darknet/YOLO discord server so I can fix or better document whatever you didn’t understand: https://discord.gg/zSq8rtW
Note the recommended training command on the repo and in the Darknet/YOLO FAQ (https://www.ccoderun.ca/programming/yolo_faq/#training_command) is not exactly the command you state.
- You left out the very critical parameter
-map - You wouldn’t normally specify the weights file
yolo-obj_2000.weightsunless you are resuming training from the 2000th iteration.
Make sure you understand the license difference between v8-9-10, and the completely free and open-source license of Darknet/YOLO.
Lastly, you may want to see this recent video which compares some of the more recent versions of YOLO against the original Darknet/YOLO: https://www.youtube.com/watch?v=2Mq23LFv1aM
Disclaimer: I maintain the Darknet/YOLO codebase. I’m the author of tools such as DarkHelp and DarkMark.