This collection of early works represent Mital’s explorations of image and video synthesis, primarily throughout the course of his doctoral studies and the development of his thesis on Computational Audiovisual Scene Synthesis. In this series, Mital reconstructs diverse visual content ranging from personal memories to iconic art and presents outputs ranging from still and video artworks to procedural augmented reality experiences.
Nanital | 2013
Footage from Mital’s parents’ honeymoon following their arranged marriage are generatively collaged with the colors of Indian wedding ceremonies.
Visual Synthesis Experiments | 2013
Early experiments in Mital’s process for creating visual syntheses. These experiments can be seen as early sketches in form, exploring texture, color, and different palettes.
Augmented Reality | 2013
One of the earliest video examples using a procedural augmented reality. In this video, a Vuzix Wrap 920AR augmented reality device is used to stream the video content from 2 head mounted cameras attached in front of the screen being displayed to each eye. The resulting images are shown here for one eye only. As the algorithm has no semantic knowledge, the resulting syntheses cannot differentiate different letters on the keyboard, though still manage to use its proto-object representations to collage together a single key across an entire keyboard.
Realtime Visual Smash Up | 2012
Realtime Visual Smash Up presents a system for realtime mosaicing of visual content. The system learns objects from the camera image and displays the resulting visual mosaic. Only unique objects are learned creating a hallucinatory resynthesis of the original camera image. For instance, all keyboard letters are replaced by only a single letter since based on purely visual appearance, a single letter is unique enough to portray the keyboard.
Mona Lisa vs. Arcimboldo | 2012
The works of the 16th century Mannerist painter Giuseppe Arcimboldo are shaped to resemble one of the most iconic portraits in history, Leonardo da Vinci’s Mona Lisa. Arcimboldo, known for his composite head portraits which mold faces from fruits, vegetables, flowers, and other objects, defied the figurative conventions of his time, deconstructing their very essence into an outlandish representation of a portrait. In a 16th century head nod to the painter, Mital extracts and deconstructs the same elemental proto-objects, redefining them to resemble the timeless Mona Lisa.
Harry Smith vs. Pink Elephants | 2011
A perceptual model based on proto-objects is presented as a visual reconstruction of Harry Smith’s Early Abstractions. Mital trained the model on a scene from Dumbo, Pink Elephants, asking it to interpret Harry Smith, having only knowledge of Dumbo. The reconstruction is surprisingly able to capture a wide variety of the abstract images and movements in Harry Smith, as well as capture after-image. This model is an early prototype of Mital’s PhD work on visual resynthesis.
The Simpsons vs. Family Guy | 2011
This work represents the development of a method for resynthesizing existing videos using material from any other video(s). This process starts by learning a database of objects that appear in the set of videos to synthesize from. The target video to resynthesize is then broken into objects in a similar manner, but also matched to objects in the database.
Dimensions | 2011
Regurgitating Svankmajer’s “Dimensions of Dialogues” after having eaten Svankmajer’s “Food”. This is a video resynthesis which first begins by learning a large database of objects by segmenting each frame of Svankmajer’s “Food”. Having digested this video into a large database, the resynthesis of Jan Svankmajer’s “Dimensions of Dialogue” (shown on the left) is done by matching similar objects in this video to the ones in its database, creating a mosaicing only using material from “Food”. Made with openFrameworks and OpenCV.