Have a UROP opening you would like to submit?
Please fill out the form.
MAS: Media Arts and Sciences
Jan 1, 2020
Use neural-art to render news broadcasts as if they were painted by artists rather than laid out by set designers.. We have a set of news snippets from DirectTV, and ideas to create news sets inspired by impressionist artists. We also have scripts that can run on GPU-CUDA machines. The following tasks need to be accomplished (Note: scripts are available for all portions of the task, but may need minor tweaking sometimes): Phase 1 * Get set up on the Necsys GPU machines, at the media lab * Break up news videos into segments we will run through the art script, i.e segments of just the news set + news caster, and the remaining pieces of the news video. * Segment the news set snippets into the foreground and the background, i.e segment the videos into the news set and the news caster. * Finding impressionist art, that can be used to modify the videos * Understand the balance between placing an importance on the news content vs the art style, in the neural net. * Convert the news video snippets into aesthetically pleasing videos using pre-existing scripts, where some minor modification to the script may be necessary. * Mash up the news segments that were split up earlier to make a single news video, with the art news set and the associated video segments. Phase 2 * Modify the script to generate the background every N frames, instead of every frame. * Play with not just the news set and news caster, but also segmenting objects in other frames of the news.
Experience with general programming/scripting in python Experience with machine learning/neural networks Familiarity with Docker preferred