This repository contains the code for the reproducibility of the experiments presented in the paper "A Deep Learning Approach for Imputing Sparse Spatiotemporal Data".
Authors: Kehui Yao
The architecture of the ST-Transformer is depicted in the following figure. The model consists of three main components: an input encoder, a spatiotemporal transformer encoder, and an output layer.
Results on HealingMnist:
Method | MSE |
---|---|
ST-Transformer | 0.054 +/- 0.000 |
GP-VAE | 0.114 +/- 0.002 |
Interpolation | 0.135 +/- 0.002 |
Mean | 0.210 +/- 0.004 |