One dataset for Testing after merging with another

Community Help
1

I am using my own annotated dataset together with another public available dataset to improve the number of cases. Since my dataset is more important in my study I would like to use only this one for tests and maybe validation as well. The documentation of Roboflow tell me that this should be possible by uploading a new set of data. This is not shown when I click on “rebalance train/test split” when I use “generate new Version”
Also: Is there a an easier way to only use one of the merged sets for Test/Validation

This category is for question related to accounts on https://app.roboflow.com/

  1. Object Detection of Road Damages
  2. MacOS, Chrome Version100.0.4896.127
2

Hi David, you’ll want to click Generate New Version and then click on “2” within the UI to change the train/valid/test split.

From here, adjust the sliders so that train reads “0%” and valid and test match the values you are looking for. Continue on to generate the dataset version with no preprocessing or augmentation steps other than “Auto Orient.” This is to save a “raw images” version of the test/valid set.

Once the version is generated, export in COCO JSON and add to a new project with the Existing Train/Valid/Test split.

You can add new images to the dataset as “Add all images to training set” or any of the other available settings.

This should result in the split you were looking for from what I understood.

1 Like
3

Thank you so much for the great response, that is exactly what I was looking for

4

You’re welcome, I’m happy to help!

1 Like