-
This is a web application for custom multi-class/multi-object annotation. Python Flask web framework was used to receive annotation request from frond-end.
-
The front-end web page is implemented through HTML, Javascript, and jQuery, the front-end supported functions include:
- customized class/object definition
- toolkit to specify user interactive tool (pen, rectangle, back-end algorithm selection)
- history panel to undo unwanted operation, etc.
- canvas for user to interact with the back-end service.
- image gallary and file I/O support.
-
On the back-end, Python Flask web framework was used to receive annotation request from front-end, and implemented image processing functions include:
- Deep Object selection algorithm, that generate initial object mask from pos/neg strokes under Tensorflow 1.12.0
- GrabCut to refine object mask based on RGB color and the DL object mask
- Generated mask mush connect to the given positive strokes.
- Lock/unlock the previous annotated objects so new generated mask have the first/lowest priority.
- manual mode that what the user draws are what the final object mask get.
-
Zheng Zhou
-
Ali Behnoudfar
-
Nihar Ashesh Doshi
-
Damanpreet Kaur
-
Nicholas Kiddle
-
co-workers (time ordered):
- Pierce, Brett Richard
- Mewalal, Ritesh
- Peremyslova, Ekaterina
- Nagle, Michael F
- Strauss, Steven