Define a test set

Hello to all,
This is my first post here and I’m still exploring things, and I’m missing something obvious: how to define a test set once the image has been uploaded?
They show up in the Assign tab but they are not meant to be labeled since they are test images.

I did find the option to define train/val/test portions/set on the generate tab but I’ve uploaded the Test/val images separately, so now how can I put the other images in the dataset and define them as test set?

Hi @diramazioni - on Roboflow, you will want to label the test images, actually.

The reason for this is because we provide the ability to compare Ground Truth labels to Model Predictions on your Valid and Test sets with the Visualize feature that is available on versions with a trained model / model weights.

If you want to ignore this option, then you can take the images you want to use for the test set, and upload them to a separate project instead, to later export and “merge”, or include, with the remaining dataset images.

I see. since the test images has not been annotated, can I use the model I’ve trained so far in colab with label assist? I cannot seems to find a way to import the weights

Yes, but it depends on the model you’ve trained, and project type.

We currently offer model weights uploads from custom trained models for these architectures:

  1. YOLOv8 Object Detection
  2. YOLOv5 Object Detection
  3. YOLOv7 Instance Segmentation

I’m making a YOLOv8 Object Detection project, and following the notebook you have provided at the end it fails to deploy because it says another version is present (even if I never reached this stage before)
On the web interface I’ve then tried to click deploy but that got stuck and after several hours I still see the spinning circle.
What should I do?

Hi @diramazioni - please see this post for instructions on how to remedy this: ANNOUNCEMENT: Custom Model Weights Upload Overview

You’ll need to generate a new version. We’re working on a stable fix for the issue as well.