”Subject Extraction for Scene Reconstruction in VFX” by Vincent Dorenkamp at Blender Conference 2026

Most VFX workflows start with the reconstruction of the scene based on the footage, in order to integrate new CG assets with the existing geometry and lighting of the scene. During that process an important step is the separation of the subject(s) from the background.

In this talk Vincent will showcase a custom machine learning model for extracting subjects from footage for a better cg integration workflow.
We’ll look at the current state-of-the-art in image matting, this new model’s architecture, color reconstruction for half-transparent subjects and how to work safely with external libraries in extensions asynchronously.

–

Learn more about Blender Conference 2026 at https://conference.blender.org/2026/

#b3d #BCON26