Core Concepts
提案された方法は、不利な気象条件下での意味的セグメンテーションマップの推論を可能にし、言語ガイダンスを活用してモデルの性能向上を実現します。
Abstract
Abstract:
Existing models show performance drop under adverse weather conditions.
Proposal of WeatherProof dataset with clear and adverse weather image pairs.
Introduction:
Semantic segmentation importance in various applications.
Performance degradation of models under adverse conditions.
Adverse Image Fine-tuning Language Guidance:
Introduction of WeatherProof dataset with improved mIoU using language guidance.
Data and Analysis:
Weather effects combinations impact model performance.
Methods:
CLIP Injection Layer introduced to improve model's resilience to weather conditions.
Experiments:
Improved mIoU on WeatherProof and ACDC datasets using language guidance.
Conclusion:
Proposed method shows significant performance improvements but has limitations.
Stats
存在するモデルは、不利な気象条件下で性能が低下することが示されています。
提案されたWeatherProofデータセットでは、言語ガイダンスを使用してmIoUが向上しています。
Quotes
"By leveraging CLIP-based language guidance, our models perform up to 10.2% better on our WeatherProof test set."
"Our method improves upon standard training techniques by up to 8.44% in mIoU on the widely used ACDC dataset."