Hey!
Could we please have a point dropper tool for image annotation?
We need annotators to accurately mark very speficic parts of a document image, and it would be great to do this natively without any workarounds.
Hi @tuomo_h:
Thanks for the suggestion! To better understand your use case: are you thinking of keypoint annotation as used in computer vision? That means the annotator clicks on the image to drop a labeled point on a specific spot, for example facial landmarks or body joints. Points can also be linked to each other and ordered, for example the joints of a skeleton or the corners of a table.
Is that close to what you have in mind? Also, would linking and ordering points be useful for your documents, or are single labeled points enough?
Would you need to zoom in and pan around the image to place points accurately?
Linking/ordering and zooming/panning would be separate features from a basic point tool, so it helps us to understand how common these needs are across different use cases.
Hi @magdaaniol!
Regarding our use case, think of a multimodal co-reference resolution task, which is extended over text, photographs, diagrams, etc. The user will drop points across these references, which make up a chain.
More generally, I think it would be good to have two modes for the point dropper tool:
- Individual points: there is no connection or order between the points
- Linked points: one can connect the points to each other using lines
I think a zoom and pan feature would be super useful for our use case as well, although having the ability to hide pre-existing annotations helps already.