Artificial Intelligence Stack Exchange is a question and answer site for people interested in conceptual questions about life and challenges in a world where "cognitive" functions can be mimicked in purely digital environment. It only takes a minute to sign up.
Sign up to join this community
I want to train a convolutional neural network for object detection (say YOLO) to detect faces. Consider this image:
In this training image, I have many people, but only 2 of them are annotated. Is having this kind of images (where target classes are not all annotated) will train the network to ignore positives?
If yes, are there any techniques to solve the issue apart from annotating the data (I don't have enough resources to do that).
The neural network will learn what we teach it, for example with that image only, after finish training, your model will hard to recognize humans with dark skin, glasses, big eyes, etc, the features that two annotated targets don't have.
If your data is big enough, and contain all the feature of humans face, the result should be good.
If not, I recommend a semi unsupervised learning method which called Noisy Student. Quick explanation, you take a part of data, add noise (augmentation, drop-out, stochastic depth), and train. Then use that model to label the rest of the dataset and train a new bigger teacher (such as YOLOv3 > YOLOv1) and repeat it. They told that this method was even better than you labeled all the dataset.
You will need to choose the data to train the teacher really carefully, then pray for the result.
