Year: 2025
Type: conference-paper
Abstract: This paper outlines our solution to the WWW25 Text-based Person Anomaly Search Challenge, structured into two primary strategies: 1) Continuous pre-training of existing models, and 2) Model ensemble techniques. Initially, we adapt open-source pre-trained models such as EVA-CLIP [4] and OPENCLIP [3] through continuous training on the specific competition datasets. We then implement two model ensemble strategies to refine prediction accuracy. Our approach culminates in achieving Recall@1, Recall@5, and Recall@10 values of 89.28%, 99.34%, and 99.75% on the challenge's online leaderboard.
Institutions: Beijing Institute of Technology, Northeastern University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4410636816