Keras: Visualizing image summaries from intermediate layers using Tensorboard
Is there a way to use keras to visualize image summaries using tensorboard? The tensorboard callback currently doesn't seem to support them. If there is no existing functionality, can someone give an example of implementing a separate callback for this (or any other suitable way to achieve this)?
hj1210
>All comments
Recently ran across this issue.
The TensorBoard callback creates a self.merged layer with nodes comprised of the histograms and images, but it only attempts to write this summary info if model.validation_data is defined. If you are using fit_generator with a generator for the validation data, validation_data is set to None, and therefore the images and histograms are never written.
In addition, the code that generates the merged layer looks like it needs to be updated... I'm getting all kinds of invalid size errors related to the image nodes.
At a minimum this should be documented, and certainly the class could use some improvements.
algoterranean
on 23 Jan 2017
Related issues
KeironO
路
3Comments
harishkrishnav
路
3Comments
braingineer
路
3Comments
somewacko
路
3Comments
MarkVdBergh
路
3Comments
Most helpful comment
Recently ran across this issue.
The
TensorBoardcallback creates aself.mergedlayer with nodes comprised of the histograms and images, but it only attempts to write this summary info ifmodel.validation_datais defined. If you are usingfit_generatorwith a generator for the validation data,validation_datais set to None, and therefore the images and histograms are never written.In addition, the code that generates the merged layer looks like it needs to be updated... I'm getting all kinds of invalid size errors related to the image nodes.
At a minimum this should be documented, and certainly the class could use some improvements.