09 Jul 2025
StixelNExT++ introduces a lightweight, end-to-end neural network for 3D monocular scene segmentation and representation, achieving real-time inference at 10 ms per frame and a 63.8% F1-Score for holistic obstacle detection. The system generates a compressed 3D Stixel World from a single RGB image, suitable for collective perception by leveraging a depth-aware loss and non-linear depth discretization.
There are no more papers matching your filters at the moment.
