Is there any way to crop already-labeled images into multiple parts, preserving the labels using Roboflow?


Roboflow has a pretty useful crop function in preprocessing tools that lets you crop a labeled image while preserving the labels; is there any way to use this to break a labeled image into multiple parts? For example, I have 10 images and want to break each of them into 10 parts, resulting in 100 images, while keeping all the labels’ positions intact.


Have you tried the tiling preprocessing step? Launch: Edge Tiling During Inference

1 Like

Thanks, this was really helpful. Is there any way to regenerate the original images, with original labels? (i.e. combining labeled broken parts of objects into a single labeled original object)

If you generated a Version without tiling, you can go back to that Version download the data

I think that should help accomplish what you need but please let me know if not.

Hi sorry,

Thanks for trying to help.

So right now if I have 10 images and I tile each into 3x3, I’ll have 90 images after tiling. I can split these into train-val-test and I can achieve pretty good results on the test set after training. However, my original goal was to detect objects in the original untiled image. Is there a way to automatically visualize the original untiled image with all the labels and stuff that my model generated?

Sorry, just’d like to bump this

You can use the Deploy options (Deploying Your Computer Vision Model with Roboflow) to run your original image through the trained model. Or reference the Docs for all deployment options Supported Deployment Devices - Roboflow Docs

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.