HG-Lane: High-Fidelity Generation of Lane Scenes under Adverse Weather and Lighting Conditions without Re-annotation

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• HG-Lane improves the overall mF1 score of CLRNet lane detection by 20.87% on their benchmark.result 95%
• HG-Lane generates lane scenes under adverse weather and lighting conditions without requiring re-annotation.method 90%
• The framework constructs a benchmark containing 30,000 images of adverse weather and lighting scenarios.technical 90%
• Existing lane detection datasets like CULane and TuSimple have limited data under extreme weather conditions.limitation 85%
• Models trained on existing datasets become unreliable in adverse weather, potentially leading to safety-critical failures.limitation 80%
• HG-Lane uses a dual-stage, control-guided diffusion framework with pre-trained models like ControlNet and InstructPix2Pix.technical 85%
• The method significantly improves F1@50 scores across multiple adverse condition categories, with the largest improvement (38.8%) for snow conditions.result 90%
• Reliance on synthetic data might not capture all edge cases in real-world conditions.limitation 75%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card