https://phuchoang2603.github.io/sis-arxiv-vad-papers/authors/gijs-dubbelman/Recent content in Gijs Dubbelman on sis-arxiv-vad-papersHugo -- gohugo.ioenSun, 01 Oct 2023 00:00:00 +0000https://phuchoang2603.github.io/sis-arxiv-vad-papers/papers/simplifying-traffic-anomaly-detection-with-video-foundation-models/Sun, 01 Oct 2023 00:00:00 +0000https://phuchoang2603.github.io/sis-arxiv-vad-papers/papers/simplifying-traffic-anomaly-detection-with-video-foundation-models/The paper investigates the use of simple encoder-only Video Vision Transformers (Video ViTs) with various pre-training strategies for traffic anomaly detection (TAD), demonstrating that with strong pretraining and domain adaptation, minimal architectural complexity can outperform complex prior methods, highlighting the importance of pretraining strategies like Masked Video Modeling (MVM).