How to distinguish training vs validation loss from tensorboard

[cute-tiger]

When use the" tensorflow object detection api "to training a detection model, and use the recommended directory structure to run tensorboard,command line is tensorboard --logdir=${MODEL_DIR}.

it's output is this image:it have many loss function. Are these loss from training or validation process? If these are from training process only, how to check validation loss using tensorboard?

[divyanshusharma1709]

If you're using the latest TensorFlow Object Detection API (using model_main.py and not train.py), then the Validation Loss can be found under 'loss' and the Training Loss is under 'loss_1' or 'loss_2'.

[tispratik]

Do we have to pass a special validation flag or a file for it to separate out the cross validation data? In our case, even though the validation loss is low, we don't see good generalization.

[ojasvisancheti]

@divyanshusharma1709 In my case only the loss tag is present, no Loss1 or Loss2, Is i am missing something? please help

[kabyanil]

I'm not seeing loss_1 and loss_2. I'm only getting a tab called loss.

what am i missing? is there any parameter required to pass in order to get both train and validation loss in tensorboard?

[markusbink]

I'm having the same issue (using model_main.py) that i have multiple losses (i.e. classification loss, localization loss, total loss etc.) but no two curves as @divyanshusharma1709.