Table 2.

Summary of the hyper-parameters used to train the multi-modal network. Hyper-parameters were tuned heuristically.

Hyperparameter	Value
Learning rate	$1·{10}^{-4}$
Optimizer	Adam (${\beta }_1=0.9,{\beta }_2=0.999$)
Batch size	16
Dropout	0.50
Epochs (frozen backbone)	50
Epochs (fine-tuning backbone)	50
Loss function	Mean Squared Error
Image dimensions	(1024, 1024)
Timeseries length	4223 timestamps