End-to-End Computer Vision Driven by Python: Research, Education, and Practical Innovation

Abstract

Nowadays, with the development of artificial intelligence technology, image processing is widely used in many areas such as industry or education field. Python is becoming the de facto standard of choice when developing computer vision software mainly due to its cross platform capability, simple programming style as well as a rich set of open source third party libraries such as OpenCV [1], TensorFlow [2], and PyTorch [3]. Unfortunately, today’s programming practices require tedious coding and cumbersome task coordination and a disconnect between teaching methods and real-world application settings. In this work, we introduce the EECVF vision system by using visual programming methods for providing an intuitive, easy-to-use interface. By dividing the program flow to independent working processes and offering simplified operation steps the model facilitates efficient data-driven control of processes; it simplifies the end-to-end workflow starting with data management and design of algorithms through deployment; It lowers the technological barrier of implementation so that it is more feasible to implement as an assignment for university courses; and enables us to perform further studies with higher fidelity methods like MD simulations or explainable ML, while having many shortcomings for example lack of interactivity and dependence on specific computers; future releases will remedy these problems. In the future we plan to deploy this platform using a server in the cloud and load balancing techniques and the use of GPUs. Given its generality, it is likely that this system could be used more widely where we expect to see a growing interest and applications for example in the following fields.