An AI Based Enhanced and Customized Doodler-Bot with Vector Mutation And Clustering

Abstract

Currently high usage of smart phone is observed in millennials for learning, playing games or drawing. Drawing being quite popular among all age groups of people, still the most generic problem faced is the fear to draw neatly and correctly. We have tried to make drawing experience more innovative and effective by proposing a unique AI based Doodler-Bot which will help people of all age groups to improve their drawing skills. There exists some way through which the user's sketch can be improved by replacing it with the professionally drawn sketch stored in database, but by doing this the user's originality of sketch is lost. Our proposed work will help user in drawing better sketches by generating new and unique sketches which are better version of user's sketch. Whenever user draws a sketch, the application would first predict the class of sketch and then the generator model will be used to generate new sketches depending on user's sketch and display all similar suggestions to user as output. The results are improvised in comparison to existing Sketch RNN model. First, by modifying the model architecture i.e. by mutating the latent space of user sketch with similar latent space of sketches available in cluster data. Secondly, the model is also trained on improvised dataset (Quick draw dataset and Newly generated dataset). New dataset is generated by converting the SVG format images to the stroke format as required by model for training. The proposed AI based Doodler-Bot will intelligently guide and improve the drawing skills of the person.