You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi! I am plotting some custom stuff with pyplot during training and want to add these plots to TensorBoard as images. Since TensorBoard uses PNG, it would be really convenient to have add_image which accepts a file (or BytesIO) as input. It would be mush easier to save a plot to memory in PNG format and skip all that tensor juggling.
The text was updated successfully, but these errors were encountered:
HI, The next release (tensorboardX) will also support numpy array as input. I think it's better to let user handle the image loading things. Users only need to define a function to load whatever format they want and feed the numpy array to add_image. As for skipping file IO, do you use method like this: https://stackoverflow.com/questions/43363388/how-to-save-a-greyscale-matplotlib-plot-to-numpy-array or are there better ways?
This request might be now reconsidered in light of the add_figure support. There's now a pattern of providing convenience functions that rely on specialized encoding. It seems these conveniences are implemented at the SummaryWriter level, which keeps them out of the lower layers.
Hi! I am plotting some custom stuff with pyplot during training and want to add these plots to TensorBoard as images. Since TensorBoard uses PNG, it would be really convenient to have
add_image
which accepts a file (or BytesIO) as input. It would be mush easier to save a plot to memory in PNG format and skip all that tensor juggling.The text was updated successfully, but these errors were encountered: