4. Example dataset
We can use the dataset that we used in Chapter 11, Object Detection. Recall that we used a small dataset comprising 1,000 640 x 480 RGB train images and 50 640 x 480 RGB test images collected using an inexpensive USB camera (A4TECH PK-635G). However, instead of labeling using bounding boxes and categories, we traced the edges of each object category using a polygon shape. We used the same dataset annotator called VGG Image Annotator (VIA) [4] to manually trace the edges and assign the following labels: 1) Water bottle, 2) Soda can, and 3) Juice can.
Figure 12.4.1 shows a sample UI of the labeling process.
Figure 12.4.1: Dataset labeling process for semantic segmentation using the VGG Image Annotator (VIA)
The VIA labeling software saves the annotation on a JSON file. For the training and test datasets, these are:
segmentation_train.json
segmentation_test.json
The polygon region stored on the JSON files could not be used as it is. Each...
