CODEWITHSUNDEEP
pythontensorflowtensorboard
golmschenk
golmschenk

Reputation: 12384

TensorBoard - Viewing image summaries side-by-side?

The network I'm working on in TensorFlow produces a pixel-wise labeling result. TensorBoard is very helpful for viewing the data during training, but if I simply add 3 image_summarys I get an layout of the images something like:

o1
o2
o3
l1
l2
l3
p1
p2
p3

(where o# is the original image, l# is the ground truth labels, and p# is the predicted labels). What I would prefer to see is a layout like:

o1 l1 p1
o2 l2 p2
o3 l3 p3

As this would make comparing the results much easier. Does TensorBoard provide any convenient way of achieving this? Or would taking all the images, mapping non-RGB ones RGB, and simply concatenating them to make one long image be the best solution? Thank you much!

Upvotes: 4

Views: 1321

Answers (2)

Sargam Modak
Sargam Modak

Reputation: 783

You can pin the images which you want to compare side by side. Below is the tensorboard images which I am using in Colab Notebook. Tensorboard screenshot

Upvotes: 0

dandelion
dandelion

Reputation: 1782

TensorBoard doesn't have solution for this at the moment; concatenating the images, like you proposed, would be a reasonable workaround.

I suggest filing a feature request on our Github :) the image dashboard is in need of some love, lots of people are using it and want improved layout and behavior.

Upvotes: 5

Related Questions