Is your feature request related to a problem? Please describe.
Now you can log only one label per pixel using the log_segmentation_image method. Sometimes we need to have more than one class per pixel (multilabel segmentation). As a workaround, we can log each label as a separate mask. But it looks like a dirty hack and a big disadvantage is that you can't display more than 1 mask simultaneously in the viewer.

Describe the solution you'd like
It'll be cool to have a method that allows logging multichannel segmentation mask tensor with binary mask in each channel. To display the intersection of different labels we can use the average color of labels in certain pixels.

Describe alternatives you've considered
As an alternative, the issue can be resolved by allowing to display of multiple segmentation masks at the same time. Of course, it's not an elegant way due to big amount of boilerplate code you should write to log multilabel segmentation mask

Additional context
Example how it looks now if we log each label as a separate segmentation mask (viewer can't display all masks together)