Lucra is a POC for the following problem statement:
->> How to reduce the time gap between hand sketch wireframe(Low Fidelity) and Design(High Fidelity) for web, mobile and any other platform? Designers on an average spend their 40% of total design stretch on this tedious task of converting hand sketched wireframe to high fidelity design
using tools like Sketch, UXpin etc. Lucra aims to reduce this span in a designer's trade drastically and help automate the process to leave space for just creative thinking.
--> The user provides a jpg/png image of the hand sketch wireframe and feeds into the matlab script.
--> There are three matlab scripts in the repo that are used to carry out this task(in order).
--> Script coordlucra2.m recognizes the different UI components coordinates down to the pixel.
--> Script lucra1.m recognizes the components class that were output from coordlucra2.m.
--> Script render2.m renders the high fidelity components(pre-defined in h1) at the exact coordinates provided by coordlucra2.m with the correct class from lucra2.m.
--> This script recognizes different ui components and places them into appropriate classes.
--> It uses Bag of features method and KNN classifier to differentiate between different components.
--> There are 23 basic components and Lucra is trained on 15(dataset-> train and eval folders).
--> Run matlab with the Computer Vision and Machine Learning Toolbox and other default ones.
--> As of 29th October 2018 Lucra is not fully automated hence the user has to run each script individually.
--> First feed a decent png/jpg of the hand sketch wireframe to the coordlucra2.m script.
--> Second use lucra1.m script for classification.
--> Finally use render2.m to produce high fidelity design of the wireframe. Choose a background or use the default one.
There is a lot of work left in Lucra like aspect ratio, padding, svg output etc. It needs to be fully automated and should yield the exact data. Moving forward it needs a much bigger dataset to have bandwidth to improvise and better ml algorithms. Still a lot left to do but atleast we have poc.