Cost Volume-based Interactive Depth Editing in Stereo Post-processing

In the post production of a stereoscopic visual media production, high-quality depth or disparity maps are essential for a number of workflow components, from initial layer separation, over editing transfers, to the final depth compositing. With automatic depth estimation not being perfect, errors lead to increased manual efforts for the artist, e.g. by requiring additional rotoscoping on the footage. We propose to fix errors in the depth maps instead of in image space, since multiple workflow steps can benefit from improved depth information. Building upon recent advances in discrete real-time stereo estimation algorithms, we guide the artist by integrating the cost volume of a stereo matching estimation into the editing parameters. Our results improve on the good results of automated algorithms and provide an opportunity for user corrections in regions that have only partially correct estimates.