Out-of-Distribution Detection and Robustness in Computer Vision

The field of computer vision is moving towards developing more robust and reliable models, particularly in the context of out-of-distribution (OOD) detection. Researchers are exploring innovative methods to improve the performance of models in real-world scenarios, where unexpected or anomalous inputs can compromise model reliability and performance. One of the key directions is the development of channel-aware and geometrical guidance-based approaches for OOD detection, which can effectively mitigate anomalous activations and enhance the separation between in-distribution and OOD data. Another important area of research is the development of data augmentation strategies, which can improve the robustness of models to different conditions, including shadows and varying camera viewpoints. Noteworthy papers in this area include TSRE, which proposes a typical set refinement method based on discriminability and activity, and SupLID, which introduces a geometrical guidance framework for OOD detection in semantic segmentation. Additionally, DriveFlow is a notable work that proposes a rectified flow adaptation method for training data enhancement in autonomous driving, and RankOOD is a rank-based OOD detection approach that achieves state-of-the-art performance on certain benchmarks.

Sources

TSRE: Channel-Aware Typical Set Refinement for Out-of-Distribution Detection

Data Augmentation Strategies for Robust Lane Marking Detection

DriveFlow: Rectified Flow Adaptation for Robust 3D Object Detection in Autonomous Driving

SupLID: Geometrical Guidance for Out-of-Distribution Detection in Semantic Segmentation

Blinking Beyond EAR: A Stable Eyelid Angle Metric for Driver Drowsiness Detection and Data Augmentation

RankOOD - Class Ranking-based Out-of-Distribution Detection

Built with on top of