Difficult to observe the loss value during training in the Terminal

[phongnhhn92]

Hi guys ! First of all, amazing work !!!
I know this is not a bug but I just want to say that it is very difficult for me to observe how the loss value changes during training. I think representing the loss values in json format is pretty bad idea. I would like to alter this for better visualization (like Keras). I will update my progress in here.

[rbgirshick]

The idea is that you can easily parse the training output as json and then plot the metrics of interest. We are unlikely to change this.

[edumucelli]

On Keras you can easily get the JSON version of the history. It will be stored in a variable for later usage as a dict, which can be easily transformed into a JSON ("to plot the metrics of interest"). However during the training the printing does not need to be pure JSON, it could be in a more readable form.