We query the histogram of the dataset and use two denoising methods to postprocess the histogram. Block-matching and 3D filtering (BM3D) tries to denoise by finding common patterns in the histogram. Graph total variation regularization (GTVR) tries to denoise using the smoothness of the histogram. We believe that common patterns and smoothness are important features for temporal data.
Place: 4th
Prize amount: $2,000
Team members: Bolin Ding, Xiaokui Xiao, Kuntai Cai, Ergute Bao