Monocular Depth Decomposition of Semi-Transparent Volume Renderings

Area of Science:

• Computer Vision
• Scientific Visualization
• Machine Learning

Background:

• Monocular depth estimation networks excel at extracting geometric information from color images.
• Applying these networks to semi-transparent volume rendered images presents challenges due to the ambiguous nature of depth in volumetric scenes.

Purpose of the Study:

• To investigate the applicability of monocular depth estimation networks to semi-transparent volume rendered images.
• To explore extensions for obtaining color and opacity information for layered scene representations.
• To evaluate different depth computations and state-of-the-art approaches for volumetric data.

Main Methods:

• Comparison of state-of-the-art monocular depth estimation approaches on volume renderings with varying opacity.
• Evaluation of different depth computation methods specific to volumetric data.
• Extension of networks to predict color and opacity for layered representations.

Main Results:

• Existing monocular depth estimation approaches can be successfully adapted for semi-transparent volume renderings.
• The adapted networks perform well across different degrees of opacity.
• A layered representation of the scene, composed of semi-transparent intervals, can be generated.

Conclusions:

• Monocular depth estimation networks are effective for analyzing semi-transparent volume rendered images.
• This adaptation offers significant potential for scientific visualization applications, including re-composition and enhanced shading.
• The layered representation aids in understanding complex volumetric data.