Stay on yolov5 or go to v8/v10?

Discussion
1

Hello,

I was considering moving my yolov5 model to v8 or v10.

Do I need to retrain a v8/v10 model all over again with the same dataset as used for my v5 model, or can I do an upgrade process on just the model to avoid that?

Thanks in advance

2 Likes
2

Retrain required. Good news is dataset format is identical so you can train directly on the same dataset with a new model:

from ultralytics import YOLO

# Load a new pretrained model
model = YOLO("yolov8n.pt")

# Train it on the same dataset
results = model.train("path/to/same/dataset.yaml")
1 Like
3

Hi Glenn,

Thank you for the info. I would probably be using the large model.

Is v8 faster to train or about the same speed as v5 please?

Can I also convert yolov8 to openvino format? - I would like to run the trained v8 model on an OAK camera.

Thanks.

4

Personally I’ve never compared training times between YOLOv5 and YOLOv8, but I’d guess that any duration differences would be relatively negligible. It is the case that YOLOv8 inference is faster than YOLOv5, so that’s a factor to consider.

YOLOv8 can be exported to OpenVino quite easily. You can view all the support export formats on the docs site page Export - Ultralytics YOLO Docs and view the OpenVino integration page OpenVINO - Ultralytics YOLO Docs

5

Great, thank you for your help.:sunglasses:

1 Like