I was wondering what the status of the python bindings for
image_transport is.
There hasn't been any serious work on Python bindings for image_transport.
They would definitely be nice to have, but actually writing them presents some thorny technical issues. Either you need to write pure-Python analogues of image_transport *and* all plugins, or find a way to wrap the C++ implementation and reuse the existing plugins. The latter approach seems more attractive in terms of code reuse and not increasing the burden on plugin implementers; but it requires changing the plugin interfaces to abstract away explicit use of roscpp (NodeHandle), as that won't play nice with rospy in the same process.
In short: this is all possible, just unpleasant and time-consuming. Since dropping in an extra republish node works adequately, I haven't made it a priority.
I've been limping along either using C++ or subscribing directly to
image messages in python, but it's starting to become a problem. The
republish idea on the wiki doesn't seem like a particularly good
solution.
Sometimes it is actually the best solution! If the image topic is published on machine A and there are multiple subscribers on a machine B, republishing the images in raw format on machine B minimizes both network bandwidth and CPU usage.
For the one-to-one case - only one subscriber on the second machine - it's true that spinning up a republish node is an inconvenience and adds a little latency. I'd suggest adding it to your subscriber-side launch file.
Cheers,
Patrick