Illumination-Guided progressive unsupervised domain adaptation for low-light instance segmentation.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Low-light conditions significantly challenge computer vision tasks due to limited photons.
• Unsupervised domain adaptation methods struggle with domain misalignment in low-light scenarios.
• Inadequate feature utilization across different stages hinders performance.

Purpose of the Study:

• To propose an Illumination-Guided Progressive Unsupervised Domain Adaptation method (IPULIS) for low-light instance segmentation.
• To progressively align features at image-, instance-, and pixel-levels between normal and low-light domains.
• To enhance feature utilization and address domain misalignment issues.

Main Methods:

• Developed an Illumination-Guided Domain Discriminator (IGD) for image-level alignment using retinex-derived illumination maps.
• Introduced a Foreground Focus Module (FFM) for instance-level alignment by integrating global and local features.
• Implemented a Contour-aware Domain Discriminator (CAD) for pixel-level alignment using contour vertex features.

Main Results:

• IPULIS achieves precise feature alignment through progressive module deployment.
• The method leads to high-quality instance segmentation in low-light environments.
• State-of-the-art performance was demonstrated on the LIS real-world low-light dataset.

Conclusions:

• IPULIS effectively addresses challenges in low-light computer vision.
• Progressive, illumination-guided feature alignment improves instance segmentation accuracy.
• The proposed method offers a robust solution for real-world low-light applications.