Monocular 3D Lane Detection for Autonomous Driving: Advancements, Challenges, and Future Directions
Monocular 3D lane detection is a crucial task for autonomous driving, enabling accurate extraction of structural and traffic information from the road in 3D space to assist in safe and comfortable path planning and motion control. Despite recent progress, there is still significant room for improvement to develop completely reliable 3D lane detection algorithms for vision-based fully autonomous driving.
Enhancing Roadside 3D Object Detection with Scene-Specific Features
A novel framework, MOSE, that leverages scene-specific features called "scene cues" to boost monocular 3D object detection performance on roadside cameras, outperforming state-of-the-art methods.
Enhancing Interpretability in Autonomous Driving through a Variational Autoencoder and Neural Circuit Policy Framework
A novel framework combining a variational autoencoder (VAE) and a neural circuit policy (NCP) to generate interpretable steering commands from input images, with an automatic latent perturbation tool to enhance the interpretability of the VAE's latent space.
Efficient and Stable Continual Test-Time Adaptation for Semantic Segmentation
A parameter-efficient model-based method named distribution-aware tuning (DAT) that adaptively selects and updates two small groups of trainable parameters to extract target domain-specific and task-relevant knowledge, effectively addressing issues of error accumulation and catastrophic forgetting during continual adaptation.