Senior Honors Thesis

Traditionally, live audiovisual performances are made behind a desk using a laptop and an audio controller. This setup restricts the performer's movements and separates the artist from their work. Moreover, it narrows what counts as the performance to the output alone — the audience is preconditioned to treat the projected image and sound as the whole event, and not the artist producing it. Cloud Tank challenges this by positioning the performer's body as the instrument, using virtual reality. As the performer, I mix spatialized audio and drive beat-aligned, reactive video layers through hand movement and hand-pose recognition. The audiovisual output is streamed and projected for the audience in real-time. Thus, the artist's movement and choreography become an integral part of the performance, adding a layer for both the audience and the artist to explore. Cloud Tank is built in Unity for the Meta Quest 3; this thesis presents its architecture, three live Pieces performed on it, and what designing the instrument revealed about embodied performance in VR.

Photographing a dimly lit scene forces a compromise: no-flash images preserve the scene's true color and atmosphere but are noisy, while flash images are sharp but ruin the ambient lighting. We reimplement the flash/no-flash framework of Petschnigg et al. from scratch and evaluate it on scenes we captured to stress-test individual stages. Beyond the core pipeline, we add the paper's continuous-flash interpolation and a guided-filter detail layer that removes the bilateral's gradient-reversal halos. We also find that the paper's literal shadow test fails under our radiometric setup, and adopt a luminance-based reformulation.