AI Paper Atlas

Research map / Point clouds / Point cloud classification

Rethinking Serialization in Linear 3D Vision: Decoupling Anisotropic Geometry from Isotropic Semantics

YinYun Yan, Liping zhang, Tingran Wang, Jiaxin Deng, Changshuo Wang, Limin Jiang, Shanwei Gao, Xin Ning

ICML 2026 · Point clouds › Point cloud classification · approach: Point-language and open-vocabulary 3D

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add ellipsoidal LAGM encoding over spherical (ablation)Sphere + MLP · S3DIS Area 5mIoU (%): 73.48 -> 74.44Improved
add GISA global aggregation over MLP (ablation)Sphere + MLP · S3DIS Area 5mIoU (%): 73.48 -> 81.94Improved
add LAGM on top of GISA (ablation)Sphere + GISA · S3DIS Area 5mIoU (%): 81.94 -> 82.62Improved
use Identity mode instead of Hilbert serialization (ablation)AnIsoNet · S3DIS Area 5mIoU (%): 74.46 -> 82.62Improved
use Identity mode instead of Morton serialization (ablation)AnIsoNet · S3DIS Area 5mIoU (%): 74.68 -> 82.62Improved
use Morton mode instead of Identity (ablation)AnIsoNet · ScanObjectNN (PB T50 RS)OA (%): 92.51 -> 94.21Improved
use Morton mode instead of Hilbert (ablation)AnIsoNet · ScanObjectNN (PB T50 RS)OA (%): 93.86 -> 94.21Improved
perturb inference input order (reverse, Morton, Hilbert, random, chunk shuffle) (ablation)AnIsoNet Identity mode · ScanNetV2 valmIoU (%): 78.47 -> 78.43No clear effect
replace PCM multi-view serialization with single-pass AnIsoNetPCM · S3DIS Area 5mIoU (%): 79.6 -> 82.62Improved
replace PTv3 with AnIsoNetPTv3 · S3DIS Area 5mIoU (%): 73.4 -> 82.62Improved
replace PTv3 with AnIsoNet without pre-trainingPTv3 · ScanNetV2 valmIoU (%): 77.5 -> 78.52Improved
AnIsoNet versus Sonata (dec.) with external pre-trainingSonata (dec.) · ScanNetV2 valmIoU (%): 79.1 -> 78.52Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy