C2P-Net: Comprehensive Depth Map to Planar Depth Conversion for Room Layout Estimation.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Room layout estimation is crucial for understanding indoor spatial configurations from images.
• Existing methods struggle with occlusions and position dependency when reconstructing indoor planes.
• Accurate reconstruction of dominant indoor planes is essential for layout estimation.

Purpose of the Study:

• To develop a robust method for room layout estimation applicable to both panoramic and perspective images.
• To overcome limitations of existing pixel-level or instance-level plane parameter learning.
• To introduce a novel Comprehensive depth map to Planar depth (C2P) conversion for improved planar depth reconstruction.

Main Methods:

• Introduced the Comprehensive depth map to Planar depth (C2P) conversion technique.
• Developed a framework that jointly learns a comprehensive depth map and planar visibility confidence.
• Proposed a novel approach for 3D layout generation via sequential planar depth map integration.

Main Results:

• The C2P conversion demonstrated applicability to both panoramic and perspective images.
• The proposed network autonomously learned planar visibility confidence through differentiable C2P conversion.
• Experimental results showed superior performance across all evaluated panoramic and perspective datasets.

Conclusions:

• The developed method offers a significant advancement in room layout estimation.
• The C2P conversion provides a more robust approach to planar depth reconstruction.
• The framework enables accurate 3D layout generation from image data.