Harmonic Feature Activation for Few-Shot Semantic Segmentation.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Few-shot semantic segmentation is challenging due to limited training data.
• Existing methods struggle with incomplete semantic understanding from support images alone.

Purpose of the Study:

• To propose Harmonic Feature Activation (HFA), a novel approach for few-shot semantic segmentation.
• To achieve dense support-to-query semantic transformation by integrating query and support image features.

Main Methods:

• HFA utilizes a bilinear model for systematic pixel-wise dense correlation between query and support images.
• A low-rank decomposition speeds up bilinear feature activation with minimal performance impact.
• Semantic diffusion is integrated to enhance global harmony and local consistency.

Main Results:

• HFA demonstrates significant improvements over state-of-the-art methods on PASCAL VOC and MS COCO datasets.
• The approach effectively addresses the limitations of conventional few-shot segmentation techniques.

Conclusions:

• Harmonic Feature Activation (HFA) offers a robust solution for few-shot semantic segmentation.
• The method enhances semantic understanding by harmonizing features from both query and support data.