HPNet: Deep Primitive Segmentation Using Hybrid Representations

Yan, Siming; Yang, Zhan-Ying; Ma, Chongyang; Huang, Haibin; Vouga, Etienne; Huang, Qixing

doi:10.1109/iccv48922.2021.00275

Cited by 23 publications

(35 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…2) Primitive fitting addresses the task of clustering input points and fitting them with geometric primitives. Standard solutions include RANSAC [38], region growing [32], supervised learning [15,24,47] and unsupervised learning [6,41]. Recently, [24] proposes an end-to-end neural network that takes point clouds as input and predicts a varying number of primitives.…”

Section: Related Workmentioning

confidence: 99%

“…Recently, [24] proposes an end-to-end neural network that takes point clouds as input and predicts a varying number of primitives. [47] further uses hybrid feature representations to separate points of different primitives.…”

Section: Related Workmentioning

confidence: 99%

“…It is natural to consider equipping machines with such part-level understanding. Over the past few years, there has been an increasing interest in 3D part segmentation tasks to support various applications [15,24,33,47] in vision, graphics, and robotics. Thanks to the availability of large-scale 3D part datasets [33,49] and the development of 3D deep learning techniques, these methods achieve impressive part segmentation results when a test shape comes from the same distribution as the training set.…”

Section: Introductionmentioning

confidence: 99%

“…A main challenge comes from the versatile cues defining parts. Shape parts can be defined via many cues like geometric primitive fitting [15,24,47], rigid motions [13,14,48], and semantic prior [28,30,33]. The same shape can be segmented into different parts based upon different cues for the seek of various applications.…”

Section: Introductionmentioning

confidence: 99%

“…For example, a modularized design has been adopted to explicitly extract motion flow and the rigid motion of local patches so that generalizable motion-based part segmentation can be achieved in [48]. Similarly, the networks are guided explicitly to extract primitive parameters in order to segment primitive parts in [24,47]. Finding such explicit part type priors requires a deep understanding of the target task, which might not always be available.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

AutoGPart: Intermediate Supervision Search for Generalizable 3D Part Segmentation

Liu¹,

Xu²,

Rao³

et al. 2022

Preprint

View full text Add to dashboard Cite

Training a generalizable 3D part segmentation network is quite challenging but of great importance in real-world applications. To tackle this problem, some works design task-specific solutions by translating human understanding of the task to machine's learning process, which faces the risk of missing the optimal strategy since machines do not necessarily understand in the exact human way. Others try to use conventional task-agnostic approaches designed for domain generalization problems with no task prior knowledge considered. To solve the above issues, we propose AutoGPart, a generic method enabling training generalizable 3D part segmentation networks with the task prior considered. AutoGPart builds a supervision space with geometric prior knowledge encoded, and lets the machine to search for the optimal supervisions from the space for a specific segmentation task automatically. Extensive experiments on three generalizable 3D part segmentation tasks are conducted to demonstrate the effectiveness and versatility of AutoGPart. We demonstrate that the performance of segmentation networks using simple backbones can be significantly improved when trained with supervisions searched by our method. Project page: https://autogpart.github.io.

show abstract