本稿では、ビデオ悪天候除去における拡散テストタイム適応手法Diff-TTAを紹介。ViWS-NetやWeatherDiffusionなど他手法を凌駕する結果を示す。
現実世界のビジョンタスクでは、予期せぬ悪天候条件が問題となる。ViWS-NetやWeatherDiffusionよりも90倍効率的なDiff-TTA手法が提案された。
Adverse weather conditions like rain, snow, and haze are common in outdoor videos. Existing models often fail to adapt to different weather conditions.
Diff-TTA integrates diffusion-based framework for video adverse weather removal with test-time adaptation, showcasing superior performance.
Experimental results demonstrate the effectiveness of Diff-TTA in restoring videos degraded by various weather conditions, both seen and unseen.
In eine andere Sprache
aus dem Quellinhalt
arxiv.org
Wichtige Erkenntnisse aus
by Yijun Yang,H... um arxiv.org 03-13-2024
https://arxiv.org/pdf/2403.07684.pdfTiefere Fragen