FlashBEV: Fast and Memory-Efficient Exact BEV Transformation with IO-Awareness
2026-07-11 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionRobotics
AI summaryⓘ
The authors focused on improving a process called Sampling-based view transformation (Sampling-VT), which helps self-driving cars understand their surroundings by combining images from multiple cameras into a bird's-eye view. They found that the usual way of doing this uses a lot of memory and is slow because it creates large temporary data. By changing how the calculations are done—doing them in a more efficient order without changing the final result—they created a new method called FlashBEV. This new method uses much less memory and runs faster, making it easier to use on devices with limited resources while keeping the same accuracy.
Bird's-eye-view perceptionView transformationSampling-based view transformationTensorized Sampling-VTGPU memory optimizationInference latencyMulti-camera fusionAutonomous drivingOperator executionMemory traffic reduction
Authors
Shunsuke Yokokawa, Hironori Kasahara
Abstract
Bird's-eye-view (BEV) perception is a core component of camera-based 3D understanding in autonomous driving, where view transformation (VT) maps multi-camera image features into a unified BEV representation. Sampling-based view transformation (Sampling-VT) is attractive because it supports dense and continuous BEV aggregation for high-resolution and long-range perception. Its deployment bottleneck, however, is systems-level: standard tensorized implementations of Sampling-VT -- which we refer to as Tensorized Sampling-VT -- explicitly materialize large height-dependent intermediate tensors, causing memory and latency costs that scale poorly with vertical resolution and the number of cameras. We revisit Tensorized Sampling-VT from an operator-execution perspective and show that it follows a gather-reduction pattern: each BEV query independently accumulates contributions across cameras and height bins, enabling thread-local accumulation with on-the-fly recomputation that eliminates the need to materialize height- and camera-dependent intermediates. Based on this insight, we propose FlashBEV, a fully fused and IO-aware execution strategy mathematically equivalent to Tensorized Sampling-VT (same operator output) while substantially reducing global memory traffic and kernel-launch overhead. Experiments show that FlashBEV achieves more than an order of magnitude lower peak GPU memory and significant inference-latency speedups, with memory effectively independent of the number of height bins, reducing the operator's peak memory to O(BCXY) (output only). This unlocks higher BEV range/resolution and vertical discretization within fixed deployment budgets on memory-constrained devices. Our contribution is an execution redesign -- same math, different execution -- that removes a key scalability barrier for deployment-ready Sampling-VT. Code available at https://github.com/yokosyun/FlashBEV