ModelNet-O

Dataset Overview

Dataset Name

• ModelNet‑O

Dataset Description

• ModelNet‑O is a large‑scale synthetic dataset specifically designed for occlusion‑aware point‑cloud classification. The dataset more accurately mirrors real‑world scenes and includes a substantial amount of data.

Dataset Characteristics

• Contains a large volume of point‑cloud data under heavy occlusion to improve the robustness of classification algorithms.
• Pre‑processing may require a long time (approximately 7‑10 days), but a downloadable pre‑processed version is provided.

Dataset Usage

• The dataset can be downloaded via the provided link and extracted to the data/ directory.
• Training and testing scripts such as train_occluded.sh and test_occluded.sh are supplied for evaluating the PointMLS method on the occluded dataset.

Related Research

• The supported research method PointMLS, based on a multi‑stage sampling strategy, achieves state‑of‑the‑art overall accuracy on the occluded point‑cloud dataset and remains competitive on the conventional ModelNet40 and ScanObjectNN datasets.

Dataset Download Link

• Pre‑processed Dataset Download

Dataset Documentation

• Data Generation Guide

Dataset License

• Apache‑2.0 License