Abstract:
Generative Image modelling is a vast area of research in computer vision that many
studies have been conducted to address such issues as an image to image translation. In
this research, the gap between the industrial designers and their production workflow
can be bridged in order to reduce the cost of the time they spend on prototyping. This
research also demonstrates an image synthesizing technique by improving an existing
technique to generate a photo-realistic image of a real-world object from a sketch.
Designing and Implementing an image generator which could generate photo realistic
images using sketch consist of several phases of development and the main stage of
development is building Generative Adversarial Network capable of translating one
form of image to another form. The main component of the implemented system can
also be defined as an image generator and this project mainly focuses implementing
and improving the efficiency of the generator.
Implementing an Image generator which could generate photo-realistic image includes
several types of tasks. One of the main tasks is integrating a new architecture which is
Nested U-Net to the generator model and manipulating it to improve the outcome of
the model. The integration of a new network architecture to the existing baseline to
improve the system in generating photo realistic images shows promising results when
compared to the existing systems the implemented system could generate images with
more accuracy and more photo realistic. The implemented system RealSketch takes a
sketch image as an input and the system tool will generate the photo realistic image
representation of the uploaded sketch image.
