Adversarial Autoencoders for Compact Representations of 3D Point Clouds

Zamorski, Maciej; Zięba, Maciej; Klukowski, Piotr; Scherer, Rafał; Kurach, Karol; Stokowiec, Wojciech; Trzciński, Tomasz Piotr

doi:10.48550/arxiv.1811.07605

Cited by 15 publications

(20 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[61] proposed using a generative adversarial network (GAN) for voxels. [1,68] proposes learning latent-GANs [37] for point clouds, while [28] proposes an autoencoding objective to stabilize the training of point cloud generative models. Similarly [15,53,46] advocate for modified generators to improve point cloud generation.…”

Section: Related Workmentioning

confidence: 99%

SurfGen: Adversarial 3D Shape Synthesis with Explicit Surface Discriminators

Luo¹,

Li²,

Zhang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Recent advances in deep generative models have led to immense progress in 3D shape synthesis. While existing models are able to synthesize shapes represented as voxels, point-clouds, or implicit functions, these methods only indirectly enforce the plausibility of the final 3D shape surface.Here we present a 3D shape synthesis framework (SurfGen) that directly applies adversarial training to the object surface. Our approach uses a differentiable spherical projection layer to capture and represent the explicit zero isosurface of an implicit 3D generator as functions defined on the unit sphere. By processing the spherical representation of 3D object surfaces with a spherical CNN in an adversarial setting, our generator can better learn the statistics of natural shape surfaces. We evaluate our model on largescale shape datasets, and demonstrate that the end-to-end trained model is capable of generating high fidelity 3D shapes with diverse topology. Code is available at https: //github.com/aluo-x/NeuralRaycaster.

show abstract

Section: Related Workmentioning

confidence: 99%

SurfGen: Adversarial 3D Shape Synthesis with Explicit Surface Discriminators

Luo¹,

Li²,

Zhang³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Point clouds are among the most popular digital representations of 3D objects, widely used in LIDARs and depth cameras. Complex perception tasks that use point clouds, such as localization or object recognition, typically treat these representations as fixed-dimensional matrices, which requires subsampling of the cloud or other pre-processing [23], [24], [25], [26], [27] .In practice, such an approach is often very limiting as the complexity of point clouds varies across object types and some objects need more points to represent their details than the others.…”

Section: Generating 3d Objectsmentioning

confidence: 99%

“…In [22] the authors introduced HyperCloud technique which use hypernetwork to produce continuous representation of 3D objects. Instead of generating object directly with the decoder [25], the HyperCloud uses parameterization of the 3D object's surface as a function transferring uniform distribution on 3D ball into surface of an object.…”

Section: Hypercloudmentioning

confidence: 99%

HyperColor: A HyperNetwork Approach for Synthesizing Auto-colored 3D Models for Game Scenes Population

Kostiuk¹,

Stachura²,

Tadeja³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Designing a 3D game scene is a tedious task that often requires a substantial amount of work. Typically, this task involves synthesis, coloring, and placement of 3D models within the game scene. To lessen this workload, we can apply machine learning to automate some aspects of the game scene development. Earlier research has already tackled automated generation of the game scene background with machine learning. However, model auto-coloring remains an underexplored problem. The automatic coloring of a 3D model is a challenging task, especially when dealing with the digital representation of a colorful, multipart object. In such a case, we have to "understand" the object's composition and coloring scheme of each part. Existing single-stage methods have their own caveats such as the need for segmentation of the object or generating individual parts that have to be assembled together to yield the final model. We address these limitations by proposing a two-stage training approach to synthesize auto-colored 3D models. In the first stage, we obtain a 3D point cloud representing a 3D object, whilst in the second stage, we assign colors to points within such cloud. Next, by leveraging the so-called triangulation trick, we generate a 3D mesh in which the surfaces are colored based on interpolation of colored points representing vertices of a given mesh triangle. This approach allows us to generate a smooth coloring scheme. Experimental evaluation shows that our two-stage approach gives better results in terms of shape reconstruction and coloring when compared to traditional single-stage techniques.

show abstract

“…This approach is refined in (Achlioptas et al, 2018) by using the PointNet architecture (Qi et al, 2017a) to encode the latent vectors. In (Zamorski et al, 2018), the focus is on improving the accuracy of generated samples by relying on an adversarial paradigm. All these approaches use fully connected layers to regress fixed size point clouds from the latent vectors, which requires millions of parameters to generate a relatively small and fixed number of points and are therefore hard to train.…”

Section: Related Workmentioning

confidence: 99%

“…As in many other areas of computer science, there are now many deep learning approaches that can process and generate such data structure efficiently. However, while recent architectures such as PointNet (Qi et al, 2017a) can handle inputs with non-fixed dimension, most cloud generative models (Achlioptas et al, 2018;Yang et al, 2018;Zamorski et al, 2018) produce outputs with a fixed number of elements. This precludes changing the topology at run-time and increasing the level of details where needed and only there.…”

Section: Introductionmentioning

confidence: 99%

NeuralSampler: Euclidean Point Cloud Auto-Encoder and Sampler

Remelli,

Baque,

Fua

2019

Preprint

View full text Add to dashboard Cite

Most algorithms that rely on deep learning-based approaches to generate 3D point sets can only produce clouds containing fixed number of points. Furthermore, they typically require large networks parameterized by many weights, which makes them hard to train.In this paper, we propose an auto-encoder architecture that can both encode and decode clouds of arbitrary size and demonstrate its effectiveness at upsampling sparse point clouds. Interestingly, we can do so using less than half as many parameters as state-of-the-art architectures while still delivering better performance. We will make our code base fully available.

show abstract

Adversarial Autoencoders for Compact Representations of 3D Point Clouds

Cited by 15 publications

References 0 publications

SurfGen: Adversarial 3D Shape Synthesis with Explicit Surface Discriminators

SurfGen: Adversarial 3D Shape Synthesis with Explicit Surface Discriminators

HyperColor: A HyperNetwork Approach for Synthesizing Auto-colored 3D Models for Game Scenes Population

NeuralSampler: Euclidean Point Cloud Auto-Encoder and Sampler

Contact Info

Product

Resources

About