I am training an object detection model in Automated ML interface with two models: yolov5 and retinanet_resnet50_fpn. The training with yolov5 is fine, the accuracy of the model is ~82%, but when I try to train the other model (retinanet_resnet50_fpn) with the same dataset an error arises saying that the loss function is exploting due to the learning_rate and sugest to decrease the learning rate to 0.0001. So I try to retrain (again) the retinanet_resnet50_fpn setting the learning_rate to 0.0001 and when I try to run the training another error arises saying "Error Input request is invalid".
I am using the azure Automated ML interface end I am setting the hyperparameter at the interface.
For this I use this documentation: https://learn.microsoft.com/en-us/azure/machine-learning/reference-automl-images-hyperparameters?view=azureml-api-2
In this doc the learning_rate hyperparameter where crtl+c crtl+v to interface to avoid typing errors, and the interface configuration was set as follow (same as the pic):
Hyperparameter name learning_rate
Distribution Fixed
Value 0.0001 #with . not ,
But when try to run the error arises: "Error Input request is invalid".
Can someone give a path to solve this, or a clue? All the others configurations were kept the same as the yolov5.
