Core Concepts
Proposing an online end-to-end OSSOD framework with semi-supervised outlier filtering and a Dual Competing OOD head to improve performance.
Stats
"Experimental results show that our method achieves state-of-the-art performance on several OSSOD benchmarks compared to existing methods."
"Our method needs only 0.62× training time and less memory."
Quotes
"We propose a semi-supervised outlier filtering strategy, which improves the OSSOD accuracy by better utilizing the unlabeled data."
"The experimental results prove the effectiveness of our DCO head."