← Downsampling Time Series for Visual Representation