AtlasNet: A Papier-Mâché Approach to Learning 3D Surface Generation

Groueix, Thibault; Fisher, M.; Kim, Vladimir G.; Russell, Bryan; Aubry, Mathieu

doi:10.48550/arxiv.1802.05384

Cited by 41 publications

(98 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Inspired by the AtlasNet [26], a manifold-based decoder module is designed to predict a complete point cloud from partial point cloud features. As shown in Fig.…”

Section: Manifold-based Decoder Modulementioning

confidence: 99%

“…1) Comparison with Existing Methods: In this subsection, we compare our method against several representative baselines that are also used for point cloud completion, including AtlasNet [26] and MSN [28]. The Oracle method means that we randomly resample 2048 points from the original surface of different YCB objects.…”

Section: A Quantitative Evaluation Of Proposed Shape Completion Networkmentioning

confidence: 99%

See 1 more Smart Citation

TransSC: Transformer-based Shape Completion for Grasp Evaluation

Chen,

Liang,

Chen

et al. 2021

Preprint

View full text Add to dashboard Cite

Currently, robotic grasping methods based on sparse partial point clouds have attained a great grasping performance on various objects while they often generate wrong grasping candidates due to the lack of geometric information on the object. In this work, we propose a novel and robust shape completion model (TransSC). This model has a transformer-based encoder to explore more point-wise features and a manifold-based decoder to exploit more object details using a partial point cloud as input. Quantitative experiments verify the effectiveness of the proposed shape completion network and demonstrate it outperforms existing methods. Besides, TransSC is integrated into a grasp evaluation network to generate a set of grasp candidates. The simulation experiment shows that TransSC improves the grasping generation result compared to the existing shape completion baselines. Furthermore, our robotic experiment shows that with TransSC the robot is more successful in grasping objects that are randomly placed on a support surface.

show abstract

“…Inspired by the AtlasNet [26], a manifold-based decoder module is designed to predict a complete point cloud from partial point cloud features. As shown in Fig.…”

Section: Manifold-based Decoder Modulementioning

confidence: 99%

Section: A Quantitative Evaluation Of Proposed Shape Completion Networkmentioning

confidence: 99%

TransSC: Transformer-based Shape Completion for Grasp Evaluation

Chen,

Liang,

Chen

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In early works, 3D volumes [13,17,58] and point clouds [46,16,45,1] are adopted as the outputs of the networks, which suffer from the problems of losing surface details or limited resolutions. With the development of the graph convolution network, many recent methods [18,23,57,27] take the triangle mesh as the output representation, most of which regress the vertices and faces directly and require initial template and fixed topology. Most recently, there has been significant work [33,40,22,12,15,8] on learning an implicit field function for surface representations, which allow more flexible output topology and network architectures.…”

Section: Related Workmentioning

confidence: 99%

“…Early work uses explicit representation, e.g. volume [13,17,58], point cloud [46,16,45,1], and mesh [18,23,57] for 3D related tasks, such as shape reconstruction, synthesis, and completion. Recently, deep implicit representation [33,40,22] shows promising performance in producing accurate geometry with appealing surface details.…”

Section: Introductionmentioning

confidence: 99%

Learning Compositional Representation for 4D Captures with Neural ODE

Jiang¹,

Zhang²,

Wei³

et al. 2021

Preprint

View full text Add to dashboard Cite

Learning based representation has become the key to the success of many computer vision systems. While many 3D representations have been proposed, it is still an unaddressed problem for how to represent a dynamically changing 3D object. In this paper, we introduce a compositional representation for 4D captures, i.e. a deforming 3D object over a temporal span, that disentangles shape, initial state, and motion respectively. Each component is represented by a latent code via a trained encoder. To model the motion, a neural Ordinary Differential Equation (ODE) is trained to update the initial state conditioned on the learned motion code, and a decoder takes the shape code and the updated pose code to reconstruct 4D captures at each time stamp. To this end, we propose an Identity Exchange Training (IET) strategy to encourage the network to learn effectively decoupling each component. Extensive experiments demonstrate that the proposed method outperforms existing stateof-the-art deep learning based methods on 4D reconstruction, and significantly improves on various tasks, including motion transfer and completion.

show abstract

“…Different supervision has been explored. Direct 3D supervision is dominant in the early works [11,18,21,23,50]. Weaker supervision that uses geometry projection (i.e.…”

Section: Related Workmentioning

confidence: 99%

Single-view Object Shape Reconstruction Using Deep Shape Prior and Silhouette

Li,

Garg,

Cai

et al. 2018

Preprint

View full text Add to dashboard Cite

3D shape reconstruction from a single image is a highly ill-posed problem. Modern deep learning based systems try to solve this problem by learning an end-to-end mapping from image to shape via a deep network. In this paper, we aim to solve this problem via an online optimization framework inspired by traditional methods. Our framework employs a deep autoencoder to learn a set of latent codes of 3D object shapes, which are fitted by a probabilistic shape prior using Gaussian Mixture Model (GMM). At inference, the shape and pose are jointly optimized guided by both image cues and deep shape prior without relying on an initialization from any trained deep nets. Surprisingly, our method achieves comparable performance to state-of-the-art methods even without training an end-to-end network, which shows a promising step in this direction.

show abstract

AtlasNet: A Papier-Mâché Approach to Learning 3D Surface Generation

Cited by 41 publications

References 0 publications

TransSC: Transformer-based Shape Completion for Grasp Evaluation

TransSC: Transformer-based Shape Completion for Grasp Evaluation

Learning Compositional Representation for 4D Captures with Neural ODE

Single-view Object Shape Reconstruction Using Deep Shape Prior and Silhouette

Contact Info

Product

Resources

About