Image annotation is a core part of object detection or segmentation. This part is the most tedious in terms of manual work in neural network development. Previously, we described three tools that are used for annotation: LebelImg, VGG Image Annotator and RectLabel. However, there are many other tools available, such as Supervisely, and Labelbox. Some of these tools perform semi-automatic annotations. The biggest challenge is creating 100,000 annotations and doing so correctly within a pixel level accuracy. If the annotation is incorrect, then the model that's developed will not be correct, and finding an incorrect annotation in 100,000 images is like finding a needle in a haystack. For large-scale project work, the annotation workflow can be divided into two categories:
- Outsource labeling work to a third party
- Automated or semi-automated...
