Semi-Supervised 3D Shape Segmentation via Self Refining.

Area of Science:

• Computer Vision
• 3D Shape Analysis
• Machine Learning

Background:

• 3D shape segmentation is vital in image processing and 3D analysis.
• Data-driven segmentation typically requires fully labeled datasets, which are costly and time-consuming to create.
• Manual face-level labeling for 3D shapes is labor-intensive.

Purpose of the Study:

• To develop an efficient semi-supervised framework for 3D shape segmentation.
• To overcome the challenge of acquiring fully labeled 3D datasets.
• To improve segmentation performance with limited labeled data.

Main Methods:

• A semi-supervised framework utilizing a small fully labeled set and a weakly labeled set with sparse scribble labels.
• An auxiliary network generates initial segmentation labels for the weakly labeled data.
• A self-refine module iteratively improves labels using the primary network's predictions.

Main Results:

• The proposed method achieves superior segmentation performance compared to existing semi-supervised approaches.
• The framework demonstrates performance comparable to fully supervised methods.
• Extensive benchmark tests validate the effectiveness of the developed method.

Conclusions:

• The semi-supervised framework significantly reduces the burden of manual labeling in 3D shape segmentation.
• The approach offers a practical solution for segmenting 3D shapes with limited labeled data.
• This method advances the field of 3D shape analysis by improving efficiency and accuracy.