Regularization Strategy for Point Cloud via Rigidly Mixed Sample

Lee, Dogyoon; Lee, Jaeha; Lee, Junhyeop; Lee, Hyeongmin; Lee, Minhyeok; Woo, Sungmin; Lee, Sangyoun

doi:10.1109/cvpr46437.2021.01564

Cited by 43 publications

(28 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our work, we develop a simple and automatic part substitution scheme for generating shapes with proper structural and geometric variations from a small number of labeled 3D shapes, whose quality is sufficient to improve network training. We also notice that recent point cloud augmentation techniques [64,65,66] that mix points of different shapes randomly to generate more varied shapes can enhance point cloud classification, and can be extended to shape segmentation [67]. However, random augmentation does not respect shape structure and can lead to limited improvements only, as our experiments show.…”

Section: Structure-aware Shape Synthesismentioning

confidence: 71%

Semi-supervised 3D shape segmentation with multilevel consistency and part substitution

Sun¹,

Yang²,

Guo³

et al. 2022

Preprint

View full text Add to dashboard Cite

The lack of fine-grained 3D shape segmentation data is the main obstacle to developing learning-based 3D segmentation techniques. We propose an effective semi-supervised method for learning 3D segmentations from a few labeled 3D shapes and a large amount of unlabeled 3D data. For the unlabeled data, we present a novel multilevel consistency loss to enforce consistency of network predictions between perturbed copies of a 3D shape at multiple levels: point-level, part-level, and hierarchical level. For the labeled data, we develop a simple yet effective part substitution scheme to augment the labeled 3D shapes with more structural variations to enhance training. Our method has been extensively validated on the task of 3D object semantic segmentation on PartNet and ShapeNetPart, and indoor scene semantic segmentation on ScanNet. It exhibits superior performance to existing semi-supervised and unsupervised pre-training 3D approaches. Our code and trained models are publicly available at https://github.com/isunchy/semi_supervised_3d_segmentation.

show abstract

Section: Structure-aware Shape Synthesismentioning

confidence: 71%

Semi-supervised 3D shape segmentation with multilevel consistency and part substitution

Sun¹,

Yang²,

Guo³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…We see this as an important step towards improving real-world 3D scene understanding. Second, inspired by MixUp [58], recent techniques propose to interpolate labels [30,59] and even input samples [8].…”

Section: Methodsmentioning

confidence: 99%

“…Yet, this is the opposite of our aim here, where we wish to keep the prediction of Y 1 agnostic to the existence of Y 2 . In particular, we preserve the complete context information of each mixed sample, whereas PointMixUp [8] distorts point clouds by interpolation and RSMix [30] only preserves the spatial structure of locally restricted chunks. We provide pseudo code in Figure 3 (right).…”

Section: Methodsmentioning

confidence: 99%

“…Mixed Sample Data Augmentation. In contrast to standard augmentations applied to single samples, MixUp pioneered by Zhang et al [58] enforces smooth label transitions by linearly interpolating between pairs of input samples and their corresponding class labels [8,21,30,48,56,58,59]. Training on these virtual training samples encourages linear behavior outside the empirical train distribution, leading the classifier to show fewer label oscillations and, thus, improved generalization [58].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Mix3D: Out-of-Context Data Augmentation for 3D Scenes

Nekrasov¹,

Schult²,

Litany³

et al. 2021

Preprint

View full text Add to dashboard Cite

We present Mix3D, a data augmentation technique for segmenting large-scale 3D scenes. Since scene context helps reasoning about object semantics, current works focus on models with large capacity and receptive fields that can fully capture the global context of an input 3D scene. However, strong contextual priors can have detrimental implications like mistaking a pedestrian crossing the street for a car. In this work, we focus on the importance of balancing global scene context and local geometry, with the goal of generalizing beyond the contextual priors in the training set. In particular, we propose a "mixing" technique which creates new training samples by combining two augmented scenes. By doing so, object instances are implicitly placed into novel out-of-context environments and therefore making it harder for models to rely on scene context alone, and instead infer semantics from local structure as well. We perform detailed analysis to understand the importance of global context, local structures and the effect of mixing scenes. In experiments, we show that models trained with Mix3D profit from a significant performance boost on indoor (ScanNet, S3DIS) and outdoor datasets (SemanticKITTI). Mix3D can be trivially used with any existing method, e.g., trained with Mix3D, MinkowskiNet outperforms all prior state-of-the-art methods by a significant margin on the ScanNet test benchmark (78.1% mIoU). Code is available at: https://nekrasov.dev/mix3d/

show abstract

“…Current data augmentation methods for point clouds can be broadly classified into heuristic methods and learning-based methods. In the former category, mixup [3,10,32] between multiple shapes is often used to synthesize novel training examples. Despite being able to provide consistent performance improvements, potentially due to stronger regularization with label smoothing, the data sample after mixup often does not resemble a realistic object.…”

Section: Introductionmentioning

confidence: 99%

On Automatic Data Augmentation for 3D Point Cloud Classification

Zhang¹,

Xu²,

Liu³

et al. 2021

Preprint

View full text Add to dashboard Cite

Data augmentation is an important technique to reduce overfitting and improve learning performance, but existing works on data augmentation for 3D point cloud data are based on heuristics. In this work, we instead propose to automatically learn a data augmentation strategy using bilevel optimization. An augmentor is designed in a similar fashion to a conditional generator and is optimized by minimizing a base model's loss on a validation set when the augmented input is used for training the model. This formulation provides a more principled way to learn data augmentation on 3D point clouds. We evaluate our approach on standard point cloud classification tasks and a more challenging setting with pose misalignment between training and validation/test sets. The proposed strategy achieves competitive performance on both tasks and we provide further insight into the augmentor's ability to learn the validation set distribution.

show abstract

Regularization Strategy for Point Cloud via Rigidly Mixed Sample

Cited by 43 publications

References 21 publications

Semi-supervised 3D shape segmentation with multilevel consistency and part substitution

Semi-supervised 3D shape segmentation with multilevel consistency and part substitution

Mix3D: Out-of-Context Data Augmentation for 3D Scenes

On Automatic Data Augmentation for 3D Point Cloud Classification

Contact Info

Product

Resources

About