Learning from Synthetic Data Using a Stacked Multichannel Autoencoder
This work addresses the challenge of using synthetic data for practical applications like photo-sketch recognition, but it appears incremental as it builds on existing autoencoder methods to bridge the synthetic gap.
The paper tackles the problem of learning from synthetic data by addressing the synthetic gap between synthetic and real data distributions, proposing a Stacked Multichannel Autoencoder (SMCAE) framework that enables more efficient learning and helps simulate real images for training classifiers, with preliminary experiments validating its effectiveness.
Learning from synthetic data has many important and practical applications. An example of application is photo-sketch recognition. Using synthetic data is challenging due to the differences in feature distributions between synthetic and real data, a phenomenon we term synthetic gap. In this paper, we investigate and formalize a general framework-Stacked Multichannel Autoencoder (SMCAE) that enables bridging the synthetic gap and learning from synthetic data more efficiently. In particular, we show that our SMCAE can not only transform and use synthetic data on the challenging face-sketch recognition task, but that it can also help simulate real images, which can be used for training classifiers for recognition. Preliminary experiments validate the effectiveness of the framework.