Revisiting Point Cloud Classification: A New Benchmark Dataset and Classification Model on Real-World Data

By Mikaela Angelina Uy, Quang-Hieu Pham, Binh-Son Hua, Dinh C. Nguyen, Sai-Kit Yeung. 2019

1,049 citations recorded by the source corpus.

Deep learning techniques for point cloud data have demonstrated great potentials in solving classical problems in 3D computer vision such as 3D object classification and segmentation. Several recent 3D object classification methods have reported state-of-the-art performance on CAD model datasets such as ModelNet40 with high accuracy (~92\%). Despite such impressive results, in this paper, we argue that object classification is still a challenging task when objects are framed with real-world settings. To prove this, we introduce ScanObjectNN, a new real-world point cloud object dataset based on scanned indoor scene data. From our comprehensive benchmark, we show that our dataset poses great challenges to existing point cloud classification techniques as objects from real-world scans are often cluttered with background and/or are partial due to occlusions. We identify three key open problems for point cloud object classification, and propose new point cloud classification neural networks that achieve state-of-the-art performance on classifying objects with cluttered background. Our dataset and code are publicly available in our project page https://hkust-vgd.github.io/scanobjectnn/.

Subfields: Computational Mechanics, Environmental Engineering, Geology, Evaluation & Benchmarks, 3D Vision & Pose

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

In Environmental Engineering

In Geology

In Evaluation & Benchmarks

In 3D Vision & Pose

Citations in the corpus

E(n) Equivariant Graph Neural Networks (2021, background): “Some examples are point clouds (Uy et al., 2019), 3D molecular structures (Ramakrishnan et al., 2014) or N-body particle simulations (Kipf et al., 2018).”
SE(3)-Transformers: 3D Roto-Translation Equivariant Attention Networks (2020, background): “Point cloud data is ubiquitous across many fields, presenting itself in diverse forms such as 3D object scans [21], molecular structures [17], or particle simulations [10].”
Geometric Back-Projection Network for Point Cloud Classification (2019, methodology): “3pre-trained models in [12], downloaded from https://github.”
Unsupervised Point Cloud Pre-training via Occlusion Completion (2020, methodology): “We use three benchmarks: Model-Net40 [55], ScanNet10 [39] and ScanObjectNN [49], the dataset statistics are summarised in Table 1.”
MVTN: Multi-View Transformation Network for 3D Shape Recognition (2020, methodology): “ScanObjectNN [30] has three main variants: object only, object with background, and the PB_T50_RS variant (hardest perturbed variant).”
PointMamba: A Simple State Space Model for Point Cloud Analysis (2024, methodology): “1, we conduct experiments on three versions of ScanObjectNN (i.e., OBJ-BG, OBJ-ONLY, and PB-T50-RS), each with increasing Table 1: Object classification on ScanObjectNN [49].”
Objaverse-XL: A Universe of 10M+ 3D Objects (2023, background): “Other datasets, such as Amazon Berkeley Objects (ABO) [25], Google Scanned Objects (GSO) [21], ScanNet3D [31], Articulated Knowledge Base-48 (AKB-48) [32] and OmniObjects3D [23] for objects, and Matterport3D [33] and Habitat-Matterport 3D [34] for scenes, improve on the diversity or quality of…”
3D-FUTURE: 3D Furniture Shape with TextURE (2020, background): “, 2016) and ScanObjectNN (Uy et al., 2019) create the datasets of 3D scans of real objects based on state of the art (SOTA) RGB-D reconstruction approaches.”
Masked Autoencoders for Point Cloud Self-supervised Learning (2022, methodology): “Motivated by this, we evaluate our pre-trained model on a challenging real-world dataset, ScanObjectNN [39], which consists of about 15,000 objects from 15 categories.”
Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training (2022, methodology): “Table 3: Shape classification on ScanObjectNN [38].”
Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline (2021, background): “Correctly classifying objects from this data is important for 3D scene understanding (Uy et al., 2019).”
PointCLIP: Point Cloud Understanding by CLIP (2021, methodology): “We experiment PointCLIP with the inter-view adapter under 1, 2, 4, 8, 16 shots also on ModelNet10 [58], ModelNet40 [58] and ScanObjectNN [52].”
Point-BERT: Pre-training 3D Point Cloud Transformers with Masked Point Modeling (2021, methodology): “While the superiority is degraded on the real-world dataset ScanObjectNN.”
CrossPoint: Self-Supervised Cross-Modal Contrastive Learning for 3D Point Cloud Understanding (2022, methodology): “ScanObjectNN [52] is more realistic and challenging 3D point cloud classification dataset which comprises of occluded objects extractred from realworld indoor scans.”
PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (2022, methodology): “5 present additive studies for the proposed training and scaling strategies in ScanObjectNN [44] and S3DIS [1], respectively.”
Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework (2022, methodology): “…et al., 2018) 69.8 73.7 PointNet++ (Qi et al., 2017b) 75.4 77.9 DGCNN (Wang et al., 2019) 73.6 78.1 PointCNN (Li et al., 2018b) 75.1 78.5 BGA-DGCNN (Uy et al., 2019) 75.7 79.7 BGA-PN++ (Uy et al., 2019) 77.5 80.2 DRNet (Qiu et al., 2021a) 78.0 80.3 GBNet (Qiu et al., 2021b) 77.8 80.5 SimpleView…”
Global-Local Bidirectional Reasoning for Unsupervised Representation Learning of 3D Point Clouds (2020, methodology): “Setups: We tested our method on ModelNet40/10 [52] and ScanObjectNN [46] benchmarks to compare with the state-of-the-arts.”