How to handle landscape and portrait images for object detection?

Community Help
1

I hav a large set annotated images. there are 40% 300x200px and 60% 200x300px.
i am trying to generate a detection model with yolo 8 and i have two questerions

Is it a good idea to scale up all pictures (landscappe and portrait) to 320x320?

is is neccesary to train two different models for 300x200 and 200x300?

kindregards fro austria

2

Hi @Helmut_Ritter, the good news here is that Roboflow makes it easy to test different approaches!

Just check out the Resize preprocessing step when you are creating a new Version. You can then experiment between two different models (using Tags and the Filter by Tag preprocessing step). These steps will allow you to train models on a subset of your data - you just need to tag the 300x200 and the 200x300 images accordingly.

3

This topic was automatically closed 10 days after the last reply. New replies are no longer allowed.