A benchmark for evaluating video inpainting models in a fine-grained, replicable manner.
Enabling temporally-consistent style transfer and inpainting via interpolation and aggregation.
Filling in missing frames by blending bidirectional predictions with a temporally-aware interpolation network.
The Moving Symbols dataset designed to expose generalization issues with existing video prediction networks.
Improving viewpoint estimation by integrating information about a single keypoint.
Automated lecture capture system that I worked on at UMass Amherst.
The project that I worked on with the UMass Center for e-Design during my REU in summer 2012.