Here is a source code example:
` plt.plot(val_fpr, val_tpr, 'b', label='Val AUC = %0.2f' % val_roc_auc)
plt.plot(test_fpr, test_tpr, 'g', label='Test AUC = %0.2f' % test_roc_auc)
plt.legend(loc='lower right')
plt.plot([0, 1], [0, 1], 'r--', label='')
plt.xlim([0, 1])
plt.ylim([0, 1])
plt.ylabel('True Positive Rate')
plt.xlabel('False Positive Rate')

task_logger.report_matplotlib_figure("Model ROC", "Validation/Test ROC AUC curve", figure=plt) `

Below is the figure which I can get if just call plt.savefig('figure.png') . And I expect the similar behavior from ClearML. As you can see there is no any legend for red line. It looks like a framework limitation if user must set legend caption for each series plot. Sure there is a workaround and we can save all figures as images but in this case we can't easily compare our series.