Toolbar & hotkeys for taking Image-level review actions

[tnc-ca-geo]

Toolbar & hotkeys for taking Image-level review actions

[nathanielrindlaub]

Notes from my conversation with Juli about beta tester feedback:

My original “label review” idea was to use arrow keys to validate/invalidate objects, and automatically iterate to the next object (tinder style). Juli brings up good point that sometimes you also want to be able to independently jump back and forth between images with the arrow keys, and not perform a validation/invalidation…

• We discussed the idea of keeping the arrow keys as-is (right/left iterates IMAGE), but having hotkeys for validation/invalidation, and having some intelligent (automated) way to change focus to objects that need review… once you validate first object it jumps to next non-validated one etc.

• Provide another keyboard shortcut to remove all tags and mark empty- "e" or something like that. (kirk)

• Provide a keyboard shortcut to open up the text box to change the current tag. For example, the current tag shows an object as an animal, but I know it is a bird, without clicking, I want to be able to hit a key like "space" to open the text box for the current object, then type "b" to see all tags starting with b, then hit "tab" to change the animal tag to bird. (kirk)

[nathanielrindlaub]

After giving this a bit more thought, I think we should abandon the idea of iterating through each Object/Label, and instead provide buttons and keyboard shortcuts for performing actions at the Image level (or rather, performing blanket actions that apply to all Objects in an Image). The main observation driving this change of heart is that the vast majority of images only contain a single Object, and for those that contain more than one Object, they're often the same species. The other observation driving this is by far the action that requires the least cognitive load right now is the big "Mark empty" button, which effectively operates at the Image level, and requires minimal thinking and mouse-eye-coordination.

The idea is that you could quickly review the most common situation (single object, single species), and provide secondary controls for more fine-grained Object and Label level interactions should you need to invalidate some objects but not all or add new Objects.

From a UI/UX perspective I think we should provide a toolbar at the bottom of all images, similar to the one Roboflow has, with controls for:

• undo / redo
• validate all / invalidate all objects
• add object
• mark empty (depends a bit on where we land with #77)
• repeat last action (i.e., if you labeled the last image with a fox, apply fox to this object) (?)
• edit label (toggle open category selector) (?)

[nathanielrindlaub]

Side note, but I also loved Roboflows tooltips for their toolbar buttons: they were compact but had really nice rich content - you get a little thumbnail gif of the action in action, the name of the tool, the hotkey/shortcut if there is one, a brief description, and a "learn more" link to the docs. Worth replicating I think.