Compared to the old workflow in FireStation/Qt the new FireStation is slower for color training and has more glitches. I do not mean the computational speed but how fast you can create and test colormaps.

Things that need consideration to be able to reach the same speed as with the old FireStation and beyond:

• upload directly on the color training widget without changing the tab
• threshold setting with mouse wheel
• undo for selection tool. This was not available in the old FS but especially now that we mark bigger and more regions more effectively because of multi-level colormaps this is something we need. I think there should be a variable number of undo steps, something like 5 should suffice
• downloading, viewing and using colormaps for online segmentation
• possibility to switch between image buffer and buffer segmented with the current robot colormap (not the local colormap, can confuse while debugging!, not only small segmented image, but ability to switch on-the-fly)
• Have the three most important images per task available for fast switching. The old FS had three small images and when clicking one it was made the primary image shown in the big buffer. This allows for really fast switching, the current switching with the host/image_id selector is too slow.

There are a few other minor issues in current FireStation:

• Start/Stop button does not properly indicate state
• the colormap reset button is too easy to reach, at least there should be a confirmation dialog to prevent accidental loss of colormap (happened several times to me, Volker noted similar "surprises")