IEEE/IAPR ICPR 2026   Oral Presentation

MAMVI: 3D Test-Time Adaptation via
Masked Multi-View Point Clouds

Inseok Kong1, Geunyoung Jung2, Jiyoung Jung2,*
1 Department of GeoInformatics, University of Seoul  •  2 Department of Artificial Intelligence, University of Seoul
* Corresponding author
MAMVI Framework Architecture Overview

Overview of the proposed MAMVI framework. MAMVI eliminates sequential optimization by unifying multi-view adaptation into a single backward pass driven by multi-view consensus, coupled with a hybrid masking strategy (fixed ratios + Beta-distributed sampling) and a confidence-based adaptive learning rate.

Abstract

3D point cloud models suffer significant performance degradation under distribution shifts caused by sensor noise, occlusions, and environmental changes. Test-time adaptation (TTA) has emerged as a practical paradigm for mitigating this issue during inference. Recently, leveraging multi-view augmentation has shown promise in improving 3D TTA performance. However, existing multi-view approaches are often constrained by sequential optimization that treats each view independently. This sequential optimization leads to substantial inference latency due to repetitive optimization steps, making real-time adaptation impractical.

To address this, we propose Masked Multi-View Test-Time Adaptation (MAMVI), which replaces sequential optimization with a unified single-step adaptation. Specifically, MAMVI utilizes a hybrid masking strategy that combines fixed ratios for stability with Beta-distributed sampling for diversity. By aggregating losses across multiple views, MAMVI performs adaptation through a single backward pass based on multi-view consensus. Additionally, a confidence-based adaptive learning rate is used to dynamically adjust the adaptation intensity for each sample. Extensive experiments on ModelNet-40C, ShapeNet-C, and ScanObjectNN-C demonstrate that MAMVI achieves state-of-the-art accuracy on ShapeNet-C and ScanObjectNN-C. Moreover, it remains competitive on ModelNet-40C while delivering 4.9–8.9 times faster inference, making it highly suitable for real-time applications.

Key Highlights & Contributions

4.9× – 8.9×
Faster Inference Speedup
SOTA
ShapeNet-C & ScanObjectNN-C
1-Step
Multi-View Consensus Backward Pass
Unified Single-Step Adaptation

Prior multi-view 3D TTA methods repeatedly optimize parameters view-by-view sequentially, causing severe latency. MAMVI unifies multi-view predictions and optimizes through a single backward pass based on consensus across views, drastically reducing adaptation overhead.

Hybrid Masking Strategy

MAMVI introduces a hybrid point masking scheme: fixed masking ratios preserve canonical geometric structures for stable consensus, while Beta-distributed stochastic sampling injects geometric diversity to prevent collapse under severe corruptions.

Confidence-Based Adaptive Learning Rate

Instead of using a uniform learning rate, MAMVI dynamically scales the step size on each sample conditioned on the model's prediction confidence, effectively preventing error accumulation on ambiguous and out-of-distribution point clouds.

Main Results

Mean accuracy under 15 corruption types

Method ModelNet-40C ShapeNet-C ScanObjectNN-C
Source Only54.1361.3536.96
TENT63.6561.0145.97
SAR64.1261.4647.26
MEMO68.1864.30–
PG-SF68.6864.4047.54
SVWA75.0465.7447.45
MAMVI (Ours)74.8066.7749.76

Top-1 accuracy (%) averaged over 15 corruption types, compared with source-free TTA methods on a Point-MAE backbone. Best and second best are marked. Per-corruption results are in the paper.

Inference throughput (FPS)

Method ModelNet-40C ShapeNet-C ScanObjectNN-C
SVWA8.998.639.03
MAMVI77.1076.6444.50
Speedup8.6×8.9×4.9×

Compared with the multi-view method SVWA, MAMVI gives similar or better accuracy at 4.9–8.9× the throughput.

Adaptive learning rate

Setting MN-40C SN-C SO-C
No ALR73.8665.6349.33
With ALR74.8066.7749.76
Gain+0.94+1.14+0.43

Confidence-based ALR improves accuracy on all three benchmarks.

BibTeX

@inproceedings{kong2026mamvi,
  title={MAMVI: 3D Test-Time Adaptation via Masked Multi-view Point Clouds},
  author={Kong, Inseok and Jung, Geunyoung and Jung, Jiyoung},
  booktitle={International Conference on Pattern Recognition},
  pages={348--362},
  year={2026},
  organization={Springer}
}