ForeverYoung

Data visualization can help to understand and analyze statistical data in a more intuitive way by graphing the data. In recent years, it is common to see companies using hand-drawn data visualization in user reports or blogs to make their content and style more relatable. This article lists some common data visualization tools and applicable charts.

Tools	rough + draw.io	matplotlib.pyplot.xkcd	chart.xkcd & cutecharts	instad.io
Scope	For existing draw.io charts, svg charts or charts that need to be drawn directly on the canvas	For data visualization charts, especially those generated by matplotlib or seaborn. Embedded in jupyter labs/notebooks	For data visualization charts with interactive requirements. Embedded in web pages or jupyter lab/notebooks	For existing svg or spreadsheet charts. Can be converted directly to hand-drawn style
Charts	Any chart, especially for direct diagrams such as flowcharts, class charts or timeline charts, etc.	Suitable for most data visualization charts, such as line, bar, pie, contour, etc.	Only supports 'bar', 'line', 'pie', 'radar', 'scatter'	Any chart, only requires DOM input format SVG or PDF charts

NVIDIA TensorRT(TRT) library is a high-performance deep learning inference engine and delivers low latency and high-throughput for deep learning inference applications. It allows users to convert the model from other popular frameworks like pytorch or tensorflow. However, TensorRT only supports float32 rather than uint8 data type as input data type, which is the most common format for image data. In this case, when we try to deploy the image-based tasks with TensorRT, it always needs to convert the images from uint8 to float32, and then transfer the float32 image date to gpu to allow the TRT engine inference the model. When the image size is large, this preprocessing stage is slightly slow. In this blog, we are trying to introduce the NVIDIA NPP library to speed up this preprocessing progress.