# Memory-efficient GPU pipelines for real-time non-line-of-sight reconstruction

arXiv: https://arxiv.org/abs/2608.28183 · HTML: https://mlsanity.com/paper/2608.28183/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 9 of its 33 references that are in the map.
- First posted on arXiv 28 Aug 2026.

## Builds on (most cited first, 9 of 9)

- [Virtual Wave Optics for Non-Line-of-Sight Imaging](https://mlsanity.com/paper/1810.07535/) (2018)
- [Non-line-of-sight Imaging](https://mlsanity.com/paper/2005.08026/) (2020)
- [Fast Back-Projection for Non-Line of Sight Reconstruction](https://mlsanity.com/paper/1703.02016/) (2017)
- [Real-time Non-line-of-Sight imaging of dynamic scenes](https://mlsanity.com/paper/2010.12737/) (2020)
- [Physics to the Rescue: Deep Non-line-of-sight Reconstruction for High-speed Imaging](https://mlsanity.com/paper/2205.01679/) (2022)
- [CPU-GPU Heterogeneous Code Acceleration of a Finite Volume Computational Fluid Dynamics Solver](https://mlsanity.com/paper/2305.18057/) (2023)
- [A comprehensive study of time-of-flight non-line-of-sight imaging](https://mlsanity.com/paper/2603.09548/) (2026)
- [3DPipe: A Pipelined GPU Framework for Scalable Generalized Spatial Join over Polyhedral Objects](https://mlsanity.com/paper/2604.19982/) (2026)
- [SET: Stream-Event-Triggered Scheduling for Efficient CUDA Graph Pipelines](https://mlsanity.com/paper/2606.05495/) (2026)

## Abstract (arXiv)

Non-line-of-sight (NLOS) imaging reconstructs scenes hidden around a corner from indirect light recorded by a single-photon avalanche diode (SPAD). A single reconstruction is a large inverse problem: billions of photon timestamps must be binned, moved through memory, transformed and inverted. As SPAD arrays raise acquisition throughput, reconstruction becomes the limiting stage. We rebuild the GPU execution of two established wave-based algorithms, f-k migration and phasor-fields, for both streaming and offline processing. On the phasor-fields side we assemble the ring-and-radius kernels of previous work once and offline, using the analytic Fourier transform of a ring, so the propagation kernel never exists in dense form at runtime, reducing the memory and bandwidth. We reorganize the pipeline of both algorithms with fused kernels, warp-level photon binning, batched transforms, CUDA graph replay, and FP16 storage applied only where it reduces the actual bottleneck. Our implementations are up to 42x faster than the reference streaming pipeline and up to 14x faster than the fastest published GPU baseline, all while using a fraction of the memory (down to 2.5%), enabling vastly larger and finer reconstructions on the same hardware, or comparable ones within a much lower memory budget. We report an ablation of each implementation choice and propose three denoising strategies enabled by the resulting frame budget for next-generation NLOS video processing.