Procedurally Generated Eye Features for Real-time Applications

Sammanfattning: Background. As computing power has increased over the years, larger parts of the digital graphics industries are moving towards using real-time rendering for their solutions. The fast iteration and visual response help make the creative processes and communication more accessible in both the desktop and smartphone markets. One noticeably growing area using real-time rendering is the communication between people and audiences using online chat applications with virtual avatars. As humans, social interaction is essential to every aspect of our health. The requirements of remote communications to satisfy the needs of social interaction are facial expressions. Facial expressions involve motion. Thus making real-time rendering mandatory for having facial expressions in computer-generated images. This thesis shows the performance result of having real-time procedural-generated eye features. Objectives. This thesis is based on an experiment by developing a data set of duration for computing different visual aspects of anime eyes, the aim is to help inform the creation of more complete and optimized shader solutions. The objectives of this thesis are to produce a profiling application, produce a collection of eye references, produce a collection of visual eye features, produce shaders based on the collections, and measure the time performance of the shaders. Method. Doing this experiment involves producing a profiling application in the C++ programming language using the OpenGL graphics API. The profiling application can render a set of shaders and measure time metrics. The experiment is executed on two different computers with different specifications. By gathering a collection of eye references from online resources, targeting popular media in the anime style, a collection of visual eye features can be established with common and unique attributes. The collections allow for the creation of a set of shaders that replicate the collected eye features. With the profiling application and shaders, a data collection of time metrics are composed. Results. The final results of the profiling experiment have a collection of 29 shaders representing different eye features. The profiling tests gave the results of how many milliseconds it took for a set number of frames to render each produced shader. This result is shown with the help of two different graphs for each computer that executed the experiment. Deconstructing the resulting data helps to analyze and distinguish interesting aspects. Conclusions. From the result of this thesis experiment, the conclusion can be made that an eye feature by itself is a fast process for the GPU to compute on its own. Aset of eye features combined to create a final eye shader including animation and eye tracking is a valid future work section regarding this thesis experiment.