Do I need train every version?

[Kaung_Khant_Zaw]

Hello… I have a project which has 1 class. I trained the class first. Then I add more images and classes and generate the new versions (which use the latest old version).

And I download the dataset with yolov8 format. Then I trained in my pc with this command

“yolo detect train data=path\data.yaml model=yolov8.yaml epochs=100 imgsz=640”

if I have new version in the roboflow and want to use that in my pc. Do I need to train from start with the same command?

And do the imgsz need to be the same with the roboflow Preprocessing resize? Like if the resize (stretch to) in roboflow is “640x640”, the imgsz in the command must “640”?

[Lenny]

Yes, if you have a new version of your dataset in Roboflow and you want to use it on your PC, you would need to download the new version and train your model from the start with the same command. This is because the new version of your dataset might contain new images and classes that your model has not seen before, and it needs to learn from these new data points to improve its performance [source].

As for the image size (imgsz), it is generally recommended to match it with the preprocessing resize value you used in Roboflow. This is to ensure that the input size of your model matches the size of the images it was trained on. However, it’s not strictly necessary, and you might still get good results even if the sizes don’t match exactly. The important thing is to make sure that the aspect ratio of your images is preserved, to avoid distorting the objects in your images [source].

Here is the command you mentioned for reference:

yolo detect train data=path\data.yaml model=yolov8.yaml epochs=100 imgsz=640

Remember to replace path\data.yaml with the path to your new dataset version, and adjust the imgsz and epochs parameters as needed.

---

Hi! I’m Roboflow’s AI powered bot. This is an automated response powered using AI with knowledge from our docs, blog, and forum. If this solves your issue, please mark it as the solution using the checkbox icon. If this doesn’t solve your issue, please follow up with what you’ve tried/why, and the community will continue to chime in as usual.

[Jacob_Witt]

Hi! You should use the same resize parameter for version creation and training.

If you are training locally, you will need to train from scratch (or set up transfer learning). You can also train with Roboflow (which has training from checkpoints).