The same concise project set highlighted on the homepage, expanded here with links, technical tags, and the historical project archive below.
Driver-level Frame Generation
NVIDIA Smooth Motion
NVIDIA
Driver-level AI frame generation
Core contributor to a driver-based AI frame generation path that infers an additional frame between rendered frames. The work focuses on making smoother gameplay available to compatible DirectX 11/12 games even when the game does not integrate DLSS Frame Generation directly.
Research and engineering around DLSS 4 Multi Frame Generation, where AI-generated frames must fit within a few milliseconds while preserving temporal stability. The work sits at the intersection of frame interpolation, optical-flow signals, GPU scheduling, and image-quality evaluation for real-time games.
Research on partition-aware post-processing and deblocking for HEVC compressed videos. The work improved reconstructed video quality, achieved about 10% bitrate saving at similar quality, and led to ICIP 2019 / IEEE TMM publications.
Compressed Video EnhancementHEVCCNNPost-processing
Built datasets and real-time deep learning systems for human fall detection and activity recognition, including a Microsoft Research Asia demo award project.
Adaptive Lossless Compression of Skeleton Sequences
Implemented lossless compression for skeleton data in surveillance video using spatial and temporal correlation, with related papers and standards proposals.
