Test data annotation och evaluation in TensorBoard

Hello! I have a question about the implementation of YOLOv5-model on custom dataset using Roboflow. When using Roboflow’s service for creating a dataset and importing that to Google colab and the model, is it the test or val data that is used when evaluating the performance and displaying it in TensorBoard? Is the annotation of the test data imported to the model as well or is that left out, meaning that we can only display the performance on test data in terms of predicted bounding boxes and not ground truth? I it possible to print the maP for the best weight? I would appreciate your help! Thank you and the team at Roboflow, what an amazing service! Regards from Sweden!

Hi @lemonelin

It’s almost certainly the validation set. A good place to review this code would be the train.py file