Notes. As a reference, the first row shows the performance of the pre-trained model used to initialize weights in the fine-tuning. From the second row forward, the first column indicates the name of the labeled dataset used to fine-tune ASTROMER. The second and third columns show the number of epochs and training time the models spend to converge. The last column is the testing RMSE evaluated on the fine-tuned (left) and pre-trained (right) models. For the fine-tuning metrics, we employ balanced testing with 100 objects per class from each labeled dataset.