Train v Test during options upload not remembering user selection

Question/Issue: I have 5000 images samples. Half of them are in test and the others are in train folders respectively. When I pre-select “Testing” or “Training” from the following list:

Upload into category

Automatically split between training and testing



the uploader just splits them anyways. I need to be able to upload them separately. Fixing with multi select only works once. When I upload the second half (“test”) to the dataset explorer, it puts most f them in the “train” category.

Additionally I am asking it to detect labels from filenames and this is not working. What do I need to do, or what are these options presented for?

Hello @alloydylan,

You have several options to upload your files and add them to either the training or testing dataset:

You can also import a info.labels file where you can specify in which dataset your want to add your data samples and your labels:

To automatically detect labels from your filename, your samples need to have names like the following:


  • rhinoceros.3456.jpg will set the label rhino for your image.
  • rose.abcd.png will set the label rose

When using the datasources, you can also use the folder name as a label see Data sources | Edge Impulse Documentation.