Martin Delorme martin.delorme6@gmail.com
(438) 274-0941

I developed a stand-alone python graphical interface using Tkinter to centralize data management and recording during motion capture shoots into a single tool. Take data, associated assets, filenames, and all pertinent details are saved in an SQL database accessed by both the tool and the post-production pipeline. The tool ensures unique take names across a project and allows assets from previous shoots to be imported during prep.

The tool connects to multiple capturing devices in order to trigger recording on all of them at the same time. It also saves each take information from these capturing devices in its database. It connects to a large range of devices: motion capture software, reference videos recorder, head mounted camera systems, DCC and game engine. It ensures integrity between assets name in its database and the connected software to remove naming disparity during the rest of the pipeline.

A web interface is also available to allow multiple users to view live shoot data as it progresses. Users can update a limited set of fields relevant to their expertise, and a comprehensive action log tracks all changes made during the shoot.

The data and reports are exported in various formats for members of the team. It is also sent to Flow PTR (ShotGrid) and the Client Selection Website with the videos. The reference videos are processed as soon as possible to allow the client to make data selection in the next hour if needed. The video compression (through FFmpeg) and data upload sections of the tool were initially combined in the main interface, but they were separated for stability and increased processing power.

I took ownership of this project when it was an unfinished, undocumented prototype with entirely custom-built components. The initial phase was bringing the base idea to a working state. The platform allows clients to review reference videos from motion capture shoots and select frame-accurate time ranges, automatically updating a project-wide "seconds budget" for character, prop, and facial assets.

The frontend features a custom, timeline-based UI where users can snap ranges together or to the playhead, bulk-edit multiple selections simultaneously and traverse the video frame by frame. This user experience continuously evolved based on direct user feedback.

The multiple cameras from the motion capture shoot are all available and can be quickly swapped from one to another while selecting time ranges. A secondary optional synchronized video player is also available for a multi-view experience. It was primarily created for the addition of head mounted camera videos (facial) to the website. There is also a 3D viewport for free-roam visualization of raw mocap data using simplified scenes and assets.

The post production pipeline can interact with the website's API to get the client's selections directly and seemlessly, ensuring up to date information.

The MotionBuilder pipeline evolved constantly to fit the current and future needs of the motion capture pipeline. Initially, the UI was purely the native UI elements from MotionBuilder. To allow for a more customizable and nicer UI, all the interfaces were transposed to PySide inserted in a dockable native UI tool.

The pipeline has access to the upstream data of the Mocap Shoot Interface and the Client Selection Website, that it uses to automate a lot of integration and exportation tasks, minimizing manual intervention.

From asset preparation to final exports, manual processes have been condensed into easy to use UIs with buttons taking care of repetitive and error-prone manipulations. This includes managing the real-time devices during the shoot, managing environment position for the motion capture shoot and for the export to engine, pairing assets with range of motions, and more.

The integration process that required manual preparation of assets per take has been automated to file selection that processes them all and leaves them ready for retargeting touch ups. Integration went from taking 5 to 15 minutes per file to 1 to 3 minutes processing time per batch of multiple files.

The files export process for delivery automatically separates the takes by client selection parts with the appropriate assets.

Throughout the pipeline, the status and assignee of tasks in Flow PTR (ShotGrid) is updated automatically when possible. For any manual step, like retargeting and animating, a UI is available with direct access to the relevent tasks for the currently opened file, increasing the ease of keeping tasks up to date.