Our R&D team specializes in creating custom deep learning solutions in Computer Vision, Sound Processing and Natural Language Processing.
Our R&D team specializes in creating custom deep learning solutions in Computer Vision, Sound Processing and Natural Language Processing.
In the realm of Computer Vision, our projects include scene comprehension for ADAS (Advanced driver-assistance systems), data synthesis and augmentation, human segmentation for mobile applications, etc.
Under the Sound and NLP categories we develop speech data synthesis (i.e Voice Conversion and Accented Speech Generation).
The majority of our research projects target scene comprehension for ADAS. Aligned with the goal of ultimately designing a system that is able to fully understand a car’s surroundings we have developed solutions for:
The scene comprehension project is customized to allow for a High performance solution, comparable to state-of-the-art results, but also a Low complexity solution that targets embedded devices for automobiles.
Data Augmentation using Generative Models
Deep generative models are algorithms built specifically for training neural networks to generate photos, videos and music, using unlabeled data - unsupervised learning. This allows us to generate “fake-but-realistic” data points from real data points - sampling (i.e. randomly generate data points) from a distribution similar to your observed (i.e. training) data. Two of the most commonly used and efficient approaches are Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN).
We are using GANs for face attribute manipulation and image domain-transfer for improving Face recognition systems robustness. Facial attribute editing aims to modify either single or multiple attributes on a face image (i.e. intrinsic properties such as color hair, age, sex, etc.). Using the aforementioned augmentation techniques, we can increase the overall accuracy of our baseline Face-recognition system, while decreasing the FAR (false acceptance rate).
Published works
Our research into scene comprehension for ADAS has led to a paper on the topic of autonomous driving that was accepted at the CVPR 2018 workshop on Autonomous Driving (“Scene Understanding Networks for Autonomous Driving based on Around View Monitoring System”[1])
[1]: https://arxiv.org/abs/1805.07029
Would you like to know more about our research? Reach out today!
