Data Visualization
Published:
When solving a general AI problem or a particular Object Detection problem, the understanding and capture of data profoundly is an important prerequisite for an effective approach to problem and model building strategy.
1. Survey the information of image dataset
One of the basic and prioritized information to capture is the size of the image. We need to know if all the images in the dataset have the same size or not.
2. Statistics on the number of bounding boxes (objects detected) in the image
The information to find out in the dataset is the number of bounding boxes in an image (appearance density). This is useful information for us to know if the problem’s data is a “crowded” form.
3. Statistics on the size of bounding boxes
The data on the size of bounding boxes give us insights into the distribution of object size in images. With these statistics, we can consider handling the “small object” problem.
For statistics, we need to show the data in the form of a chart to get a better overview.
4. Statistics of the number of objects in each class
Unbalanced data processing is not only a matter of classification problems, but also the problem of the detection tasks