Meta Analysis of Deep Learning Models for Doodle Recognition

Digital life has augmented the human horizons, code-driven systems have spread to more than half of the world’s inhabitants owing to ambient information and connectivity. Successful combination & integration of human intelligence and machine, also known as Artificial Intelligence continues to impact the future of virtually every industry. From the Harappan civilization till the 21st century, visuals have played a crucial role in our life. The use of hand-made sketches is both frequent and persistent, which empowers human beings with greater retention and the free exchange of ideas and information. This mode of communication is both effective and helpful as witnessed across various communities. Considering the noisiness of the drawings, as the illustrators are often untrained and focus on ideas rather than the technicalities of art, doodle recognition is considered to have significant consequences in computer vision, Natural Language Processing, digit recognition, pattern recognition, and speech recognition. In this paper, we aim to implement various techniques of deep learning to analyze and classify doodles into predefined labels and estimate their efficacy to eventually improve classification accuracy. We have used data from the Google Quick Draw dataset to train and test the various Deep learning Models.