Spatial-temporal Data Compression of Dynamic Vision Sensor Output with High Pixel-level Saliency using Low-precision Sparse Autoencoder.