ML-Assisted pre-labeling for VGG Visual Image Annotator

This repository contains some utils to perform ML-assisted labeling compatible with VGG Visual Image Annotator (VIA).

To do this, we need two components:

• pre-labels: the annotations we want to import into VIA
• converter: converts from binary masks to polygon annotations compatible with VIA csv format

Pre-labels

For the first point we can exploit a pre-trained model that performs binary segmentation and produces binary masks (see ml-assisted-labeling.py). This is application-specific, so you will need to either adapt the inference mode of your model or start directly from binary masks and proceed with the next step.

Converter

The binary masks are then converted into csv format compatible with VGG VIA (see VGG_VIA_annotations.py). In particular, we target segmentation applications using polygon masks.