Deep learning is a widely utilized approach speci cally for computer vision applications. Visual recognition isone of the applications utilizing deep learning. Several challenges limit the performance of visual recognitionmethods. One of the most important challenges is the insu cient number of labeled data in the datasets. Toovercome this challenge, the recent studies propose sophisticated methods which require high computationalresources, which may create another problem. That is, the implementation of such algorithms on mobile devicesis quite challenging. Especially, these issues are encountered in surveillance systems that utilize the dronesand/or CC-TVs. To solve these problems and obtain high accuracy, the network should be able to extractboth representative and discriminative features from such a small amount of data. In this paper, we proposea generative adversarial semi-supervised training method for visual recognition. Experiments are performed toevaluate a lightweight deep convolutional neural network as a classi er network that is trained by the proposedmethod and a conditional/unconditional generator networks that are examined in adversarial training.
展开▼
