Today, I’ve read a paper about anotation.
The title is "Crowdsourcing Annotations for Visual Object Detection"
.
What it is doing isn’t super new, just saying that dividing annotation work into three parts will save time as well as improve accuracy.
- Drawing: A worker draws one bounding box around one instance of the given image.
- Quality verification: A second worker verifies whether a bounding box is correctly drawn.
- Coverage verification: A third worker verifies whether all object instances have bounding boxes.
The architecture of their task division goes like this.
The results are kind of obvious, but I’ll post them.