Tensorboard logging fails with tensorboardX 1.4

[constantinpape]

The TensorboardLogger fails when logging images and using tensorboardX 1.4 with the
stack trace below.
Note that this error does not occur in tensorboardX 1.2.

  File "/home/pape/Work/software/conda/miniconda3/envs/torch41/lib/python3.6/site-packages/inferno/trainers/callbacks/logging/tensorboard.py", line 354, in log_image_or_volume_batch
    self.log_images(tag, image_list, step)
  File "/home/pape/Work/software/conda/miniconda3/envs/torch41/lib/python3.6/site-packages/inferno/trainers/callbacks/logging/tensorboard.py", line 395, in log_images
    self.writer.add_image(tag, img_tensor=image, global_step=step)
  File "/home/pape/Work/software/conda/miniconda3/envs/torch41/lib/python3.6/site-packages/tensorboardX/writer.py", line 412, in add_image
    self.file_writer.add_summary(image(tag, img_tensor), global_step, walltime)
  File "/home/pape/Work/software/conda/miniconda3/envs/torch41/lib/python3.6/site-packages/tensorboardX/summary.py", line 205, in image
    image = make_image(tensor, rescale=rescale)
  File "/home/pape/Work/software/conda/miniconda3/envs/torch41/lib/python3.6/site-packages/tensorboardX/summary.py", line 243, in make_image
    image = Image.fromarray(tensor)
  File "/home/pape/Work/software/conda/miniconda3/envs/torch41/lib/python3.6/site-packages/PIL/Image.py", line 2463, in fromarray
    raise TypeError("Cannot handle this data type")

[constantinpape]

Fixed in 1f9831f

[constantinpape]

Actually, this is only fixed for grayscale images.
I am not sure if it will work for RGB, because tbX might have changed its axis convention.
@nasimrahaman could you have a look?

[nasimrahaman]

Fixed.

[Martin-Laclaustra]

TensorboardX v1.6 solved the problem.
Maybe you should deactivate your patch, try the new version and inform tensorboardX developer.

[constantinpape]

No need to revert any changes as long as the current version is working.
As far as I remember, this error occurred because tensorboardX changed from
channel last to channel first order and we have adapted to it - this is more natural in pytorch anyways.

[constantinpape]

@Martin-Laclaustra
If you want to check yourself if the old version of the inferno logger is working with the new tensorboardX release you can check with release 0.2, which does not have the fix for the changed axis order.
https://github.com/inferno-pytorch/inferno/releases/tag/v0.2.0