SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.17720 · 2026

FlashFPS: Efficient Farthest Point Sampling for Large-Scale Point Clouds via Pruning and Caching

Yiran Chen, Cong Guo, Qinsi Wang, Yueqian Lin, Hancheng Ye, Helen Li, Yuzhe Fu, Junyao Zhang, Changchun Zhou

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 11 functions out of this paper's own repositories and ran 10 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
Yuzhe-Fu/FlashFPS canonical 10 of 11
FunctionStatusWhere it lives
concat_collate_fn Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/dataset/build.py
pointer only (licence: NOASSERTION) · get_code("65c970f35976128a")
download_and_extract_archive Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/dataset/modelnet/modelnet40_ply_2048_loader.py
pointer only (licence: NOASSERTION) · get_code("b9389712035a81db")
farthest_point_sample Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/dataset/modelnet/modelnet40_normal_resampled_loader.py
pointer only (licence: NOASSERTION) · get_code("f80066a00e7156a2")
fnv_hash_vec Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/dataset/data_util.py
pointer only (licence: NOASSERTION) · get_code("0086c892bd1df796")
is_model Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/models/registry.py
pointer only (licence: NOASSERTION) · get_code("cd2a47ded9dd84cf")
list_models Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/models/registry.py
pointer only (licence: NOASSERTION) · get_code("d14a6ca5ff9d530e")
load_data Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/dataset/modelnet/modelnet40_ply_2048_loader.py
pointer only (licence: NOASSERTION) · get_code("9ed5c671b662922a")
pc_normalize Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/dataset/modelnet/modelnet40_normal_resampled_loader.py
pointer only (licence: NONE) · get_code("4783fbece52f500e")
ravel_hash_vec Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/dataset/data_util.py
pointer only (licence: NOASSERTION) · get_code("afdc8ad15d9098b8")
register_model Ran Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/models/registry.py
pointer only (licence: NOASSERTION) · get_code("e13039fb1c347452")
download_url Not yet run Yuzhe-Fu/FlashFPS/FlashFPS-Openpoints/openpoints/dataset/data_util.py
pointer only (licence: NOASSERTION) · get_code("556e417290a6e1fd")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Point-based Neural Networks (PNNs) have become a key approach for point cloud processing. However, a core operation in these models, Farthest Point Sampling (FPS), often introduces significant inference latency, especially for large-scale processing. Despite existing CUDA-and hardware-level optimizations, FPS remains a major bottleneck due to exhaustive computations across multiple network layers in PNNs, which hinders scalability. Through systematic analysis, we identify three substantial redundancies in FPS, including unnecessary full-cloud computations, redundant late-stage iterations, and predictable inter-layer outputs that make later FPS computations avoidable. To address these, we propose FlashFPS, a hardware-agnostic, plug-and-play framework for FPS acceleration, composed of FPS-Prune and FPS-Cache. FPS-Prune introduces candidate pruning and iteration pruning to reduce redundant computations in FPS while preserving sampling quality, and FPS-Cache eliminates layer-wise redundancy via cache-and-reuse. Integrated into existing CUDA libraries and state-of-the-art PNN accelerators, FlashFPS achieves 5.16× speedup over the standard CUDA baseline on GPU and 2.69× on PNN accelerators, with negligible accuracy loss, enabling efficient and scalable PNN inference. Codes are released at https://github.com/Yuzhe-Fu/FlashFPS.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2604.17720")
get_code_for_paper("2604.17720")
have("2604.17720")

Connect an agent — have() is free.