CustomTensorboard and ValidationEvaluator callbacks

[jhaggle]

In the CustomTensorboard class you set log_dir=FLAGS.checkpoint_path + '/train'

In the ValidationEvaluator class you set validation_log_dir=FLAGS.checkpoint_path + '/val'

Can you explain how those tensorboard objects are supposed to differ? Because when I run tensorboard --logdir='path/val' and tensorboard --logdir='path/val' I get graphs with the same names, i.e. loss, pred_geo_map_loss and pred_score_map_loss. Also, the graphs don’t have the same content, se the images here: graphs. The first graph comes from the path + '/train' -directory and to second graph from the path + '/val' -directory.

(Not so important in this connection, but tensorflow board from the path + '/train' -directory also contains learning rate and small text weight graphs).

Also, how do you interpret the numbers on the axes, i.e. e.g. 0.0230 on the y-axis and 7000 on on the x-axis?

Furthermore, you are normally able to see not only training loss and validation loss, but also training ACCURACY and valuation ACCURACY. Is there now way to see this in your implementation??

[germanebr]

Hi, I know that this question is not related to your problem, but I've been having problems running the training script, and I see that you were able to run it. I was wondering if you could help me running the script, since every time that I solve an error, another one pops out immediately.