3D objects reconstruction from frontal images: an example with guitars.

Area of Science:

• Computer Vision
• 3D Reconstruction
• Geometric Modeling

Background:

• 3D object reconstruction from images is crucial for virtual environments.
• Existing methods often require multiple views or complex setups.
• Automated reconstruction from single frontal RGB images remains a challenge.

Purpose of the Study:

• To develop an automated workflow for 3D object reconstruction from single frontal RGB images.
• To adapt the workflow for various rigid object families, demonstrated with guitars.
• To enable the use of reconstructed 3D objects in immersive virtual environments.

Main Methods:

• Object detection and segmentation using convolutional neural network-based semantic segmentation.
• 3D reconstruction via warping rendered depth maps of a fitted 3D template.
• Validation using real images and ShapeNet database models, employing metrics like IoU, Chamfer Distance, and F-score.

Main Results:

• Successful automatic generation of high-quality 3D guitar reconstructions from frontal images.
• Demonstrated adaptability of the workflow to different object families.
• Quantitative evaluation confirmed the effectiveness of the segmentation and reconstruction techniques.

Conclusions:

• The proposed method offers an effective approach for automatic 3D object reconstruction from single frontal RGB images.
• The workflow is versatile and can be applied to various rigid objects.
• This technique facilitates the creation of 3D assets for immersive virtual reality applications.