Adding Background Images to Reduce False Positives
Explore how adding background images without target objects into YOLO's training set helps reduce false positives by teaching the model to better distinguish actual objects from visually similar backgrounds. Gain practical knowledge on improving model accuracy and generalization by including diverse background images in your dataset.
What are background images?
Because YOLO divides an image into a grid, all the cells that do not have any object in them are considered negative images (the left image in the example below). Let’s suppose we want to train our model on five classes: a bird, a tree, a person, a car, and a house. In YOLO, the background image usually refers to the image that does not have any of the classes that we want our model to be trained on (the right image in the example below).
Note: Background images are images that do not contain any instances ...