10. Example dataset

A small dataset made of 1,000 640 X 480 RGB train images and 50 640 X 480 RGB test images was collected using an inexpensive USB camera (A4TECH PK-635G). The dataset images were labeled using VGG Image Annotator (VIA) [5] to detect the three objects: 1) Soda can, 2) Juice can, and 3) Bottled water. Figure 11.10.1 shows a sample UI of the labeling process.

A utility script for collecting images can be found in utils/video_capture.py in the GitHub repository. The script can speed up the data collection process since it automatically captures an image every 5 seconds.

Figure 11.10.1 Dataset labeling process using VGG Image Annotator (VIA)

Data collection and labeling is a time-consuming activity. In the industry, this is typically outsourced to a third-party annotation company. The use of automatic data labeling software is another option to accelerate the data labeling task.

With this example dataset, we can now train our object detection...