I would like to know how I can label the windows or images that do not include any objects and information and how to train networks with them. These labels describe as [0 ? ? ? ? ? … ?] in Andrew’s explanation and slide.
About the labeling, for example, I have a photo that does not contain any objects, I want to make a label for this photo, the first entry is 0 which shows the picture does not contain any desired classes such as car and pedestrian. I need to know what I should put for other entries instead of ‘?’ mark when I am defining labels matrices.
On the other hand, when I want to train the model with those data, do they propagate the error for training the parameters?
In practice that’s extremely difficult, because there are a nearly infinite number of images that would not contain any of the objects you’re trying to detect. Such a training set would be either infinitely large, or hopelessly incomplete.
Labels for multiple classes are typically one-hot encoded. So if you have an image that doesn’t have any of your classes, then all the encodings would be zero.
Note that if you have a lot of images that don’t represent any of your labels, you’ll get a very skewed data set, and training it will be difficult.