System for fast and accurate visual domain adaptation

Abstract

In the field of computer vision, without sufficient labeled images, it is challenging to train an accurate model. But through visual adaptation from source to target domains, a relevant labeled dataset can help solve such problem. Many methods apply adversarial learning to diminish cross-domain distribution difference. They are able to greatly enhance the performance on target classification tasks. GAN (Generative Adversarial Networks) loss is widely used in adversarial adaptation learning methods to reduce a across-domain distribution difference. However, it becomes difficult to decline such distribution difference if generator or discriminator in GAN fails to work as expected and degrades its performance. To solve such cross-domain classification problems, an adaptation algorithm and system called as Generative Adversarial Distribution Matching (GADM) is implemented. In GADM, the objective function is improved by taking cross-domain discrepancy distance into consideration, and further minimize the difference through the competition between the generator and discriminator, thereby greatly decreasing the cross-domain distribution difference. Even when the performance of its generator or discriminator degrades, GADM is capable of decreasing the cross-domain distribution difference. The GADM algorithm and system employs a single GAN framework so as to achieve faster domain adaption with less computation resource. Specially, GADM transfers target data distribution to source one to keep accurate label dependence information, which ensures high accuracy and stability of source classifier and thus achieves better classification performance on target data.