MCEENet: Multi-Scale Context Enhancement and Edge-Assisted Network for Few-Shot Semantic Segmentation

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Few-shot semantic segmentation aims to achieve high performance with limited labeled data.
• Existing methods struggle with insufficient contextual information and poor edge segmentation.
• Addressing these limitations is crucial for advancing the field.

Purpose of the Study:

• To propose a novel network, MCEENet, for few-shot semantic segmentation.
• To enhance contextual information mining and improve edge segmentation accuracy.
• To achieve state-of-the-art performance on benchmark datasets.

Main Methods:

• MCEENet utilizes a combination of ResNet and Vision Transformer for feature extraction.
• A multi-scale context enhancement (MCE) module fuses features across scales and modalities.
• An Edge-Assisted Segmentation (EAS) module incorporates edge information for refined segmentation.

Main Results:

• MCEENet achieved 63.5% accuracy in the 1-shot setting and 64.7% in the 5-shot setting on PASCAL-5i.
• These results represent improvements of 1.4% and 0.6% over existing state-of-the-art methods.
• The proposed MCE and EAS modules effectively addressed the limitations of prior approaches.

Conclusions:

• MCEENet demonstrates superior performance in few-shot semantic segmentation.
• The integration of multi-scale context enhancement and edge-assisted segmentation is effective.
• The proposed network offers a significant advancement for tasks requiring precise segmentation with limited data.