1) Deep learning for image classification and processing:
o Deep learning based holiday image classification. Full pipeline of data collection, cleaning, preprocessing, model design, training and testing. Integrated time metadata with image feature to significantly improve the classification accuracy.
o Image super resolution, implemented CNN based image super resolution prototype, optimized for realtime mobile applications with small memory footprint and fast execution speed. Implemented generative adversarial network (GAN) based super resolution to improve the sharpness of images.
o Designed a multi-frame/video super resolution architecture by combining CNN and bi-directional LSTM.
o Designed and implemented a new loss function based on neighborhood component analysis (NCA) for feature embedding. Tested the new loss function for CNN based face verification and recognition.
2) QoS for software defined networks (SDN):
o Proposed and designed a novel QoS management framework for SDN.
o Designed and implemented a dynamic priority queue assignment and QoS routing algorithm for SDN.
o Proposed and designed a SDN based multiparty conference system.
3) Web-based realtime collaboration system
o Implemented a web based realtime collaborative system. Use WebRTC for realtime P2P video communication
o Use AR/VR technology to enhance collaborative experience, combining live media, virtual objects with enterprise business logic and process.
o Implemented a message server for session setup and teardown, using web socket for realtime interaction between clients and webGL for realtime rendering of video and virtual objects.
4) Video streaming and content delivery acceleration
o Designed a novel frame structure for camera array video streaming to achieve interactive special effects such as frozen moments and view sweeping.
o Designed TCP optimization algorithms and TCP multipath relay transmission scheme to make efficient use of available bandwidth.
