FastBEV++: Fast by Algorithm, Deployable by Design

2025-12-08

AI Summary

BEV perception faces a fundamental trade-off between high accuracy and real-time deployment on embedded automotive platforms, primarily due to the high computational cost of view transformation and its reliance on custom CUDA kernels. To address this, we propose FastBEV++, the first framework to decouple view transformation into three standard, platform-agnostic operators—Index, Gather, and Reshape—eliminating dependency on vendor-specific kernels and enabling native, high-efficiency TensorRT deployment. FastBEV++ further integrates end-to-end deep sensor fusion and temporal feature aggregation, augmented by strong data augmentation strategies. Evaluated on nuScenes, it achieves a state-of-the-art 0.359 NDS while running at 134 FPS on a Tesla T4 GPU. This work bridges the gap between accuracy and latency, establishing a new paradigm for production-ready, vision-only BEV perception systems.

Technology Category

• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Hardware-aware ML

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

The advancement of camera-only Bird's-Eye-View(BEV) perception is currently impeded by a fundamental tension between state-of-the-art performance and on-vehicle deployment tractability. This bottleneck stems from a deep-rooted dependency on computationally prohibitive view transformations and bespoke, platform-specific kernels. This paper introduces FastBEV++, a framework engineered to reconcile this tension, demonstrating that high performance and deployment efficiency can be achieved in unison via two guiding principles: Fast by Algorithm and Deployable by Design. We realize the "Deployable by Design" principle through a novel view transformation paradigm that decomposes the monolithic projection into a standard Index-Gather-Reshape pipeline. Enabled by a deterministic pre-sorting strategy, this transformation is executed entirely with elementary, operator native primitives (e.g Gather, Matrix Multiplication), which eliminates the need for specialized CUDA kernels and ensures fully TensorRT-native portability. Concurrently, our framework is "Fast by Algorithm", leveraging this decomposed structure to seamlessly integrate an end-to-end, depth-aware fusion mechanism. This jointly learned depth modulation, further bolstered by temporal aggregation and robust data augmentation, significantly enhances the geometric fidelity of the BEV representation.Empirical validation on the nuScenes benchmark corroborates the efficacy of our approach. FastBEV++ establishes a new state-of-the-art 0.359 NDS while maintaining exceptional real-time performance, exceeding 134 FPS on automotive-grade hardware (e.g Tesla T4). By offering a solution that is free of custom plugins yet highly accurate, FastBEV++ presents a mature and scalable design philosophy for production autonomous systems. The code is released at: https://github.com/ymlab/advanced-fastbev

Problem

• Resolves tension between high performance and deployment efficiency in BEV perception.
• Eliminates need for custom CUDA kernels with TensorRT-native view transformation.
• Enhances BEV geometric fidelity via depth-aware fusion and temporal aggregation.

Innovation

• Decomposes view transformation into standard Index-Gather-Reshape pipeline
• Uses deterministic pre-sorting for TensorRT-native portability without custom kernels
• Integrates end-to-end depth-aware fusion with temporal aggregation