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

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.

Vision Based Hand Gesture Recognition Using 3D Shape Context

Hand gesture recognition is a popular topic in computer vision and makes human-computer interaction more flexible and convenient.The representation of hand gestures is critical for recognition.In this paper,we propose a new method to measure the similarity between hand gestures and exploit it for hand gesture recognition.The depth maps of hand gestures captured via the Kinect sensors are used in our method,where the 3D hand shapes can be segmented from the cluttered backgrounds.To extract the pattern of salient 3D shape features,we propose a new descriptor-3D Shape Context,for 3D hand gesture representation.The 3D Shape Context information of each 3D point is obtained in multiple scales because both local shape context and global shape distribution are necessary for recognition.The description of all the 3D points constructs the hand gesture representation,and hand gesture recognition is explored via dynamic time warping algorithm.Extensive experiments are conducted on multiple benchmark datasets.The experimental results verify that the proposed method is robust to noise,articulated variations,and rigid transformations.Our method outperforms state-of-the-art methods in the comparisons of accuracy and efficiency.

Implicit Shape Reconstruction of Unorganized Points Using PDE-Based Deformable 3D Manifolds

In this work we consider the problem of shape reconstruction from an unorganized data set which has many important applications in medical imaging, scientific computing, reverse engineering and geometric modelling. The reconstructed surface is obtained by continuously deforming an initial surface following the Partial Differential Equation (PDE)-based diffusion model derived by a minimal volume-like variational formulation. The evolution is driven both by the distance from the data set and by the curvature analytically computed by it. The distance function is computed by implicit local interpolants defined in terms of radial basis functions. Space discretization of the PDE model is obtained by finite co-volume schemes and semi-implicit approach is used in time/scale. The use of a level set method for the numerical computation of the surface reconstruction allows us to handle complex geometry and even changing topology,without the need of user-interaction. Numerical examples demonstrate the ability of the proposed method to produce high quality reconstructions. Moreover, we show the effectiveness of the new approach to solve hole filling problems and Boolean operations between different data sets.

Improved quadric surfaces recognition from scanned mechanical models

This paper presents a novel algorithm for identifying quadric surfaces from scanned mechanical models. We make several important improvements over the existing variational 3D shape segmentation framework, which utilizes Lloyd＇s iteration. First, instead of using randomized initialization （which likely falls into non-optimal minimum）, the RANSAC-based initialization approach is adopted. Given a good initialization, our method converges quickly than previous approaches. Second, in order to enhance the stability and the robustness, we carefully modify the distortion-minimizing flooding algorithm by using seed regions instead of seed triangles. Third, the geometric constraints are introduced into the optimization framework. The segmentation quality is further improved. We validate the efficiency and the robustness of our proposed method on various datasets, and demonstrate that our method outperforms state-of-art approaches.

Application of optical diffraction method in designing phase plates

Continuous phase plate（CPP）,which has a function of beam shaping in laser systems,is one kind of important diffractive optics.Based on the Fourier transform of the Gerchberg-Saxton（G-S） algorithm for designing CPP,we proposed an optical diffraction method according to the real system conditions.A thin lens can complete the Fourier transform of the input signal and the inverse propagation of light can be implemented in a program.Using both of the two functions can realize the iteration process to calculate the near-field distribution of light and the far-field repeatedly,which is similar to the G-S algorithm.The results show that using the optical diffraction method can design a CPP for a complicated laser system,and make the CPP have abilities of beam shaping and phase compensation for the phase aberration of the system.The method can improve the adaptation of the phase plate in systems with phase aberrations.

A Fully Soft Bionic Grasping Device with the Properties of Segmental Bending Shape and Automatically Adjusting Grasping Range

In this paper,we propose a fully Soft Bionic Grasping Device(SBGD),which has advantages in automatically adjusting the grasping range,variable stiffness,and controllable bending shape.This device consists of soft gripper structures and a soft bionic bracket structure.We adopt the local thin-walled design in the soft gripper structures.This design improves the grippers’bending efficiency,and imitate human finger’s segmental bending function.In addition,this work also proposes a pneumatic soft bionic bracket structure,which not only can fix grippers,but also can automatically adjust the grasping space by imitating the human adjacent fingers’opening and closing movements.Due to the above advantages,the SBGD can grasp larger or smaller objects than the regular grasping devices.Particularly,to grasp small objects reliably,we further present a new Pinching Grasping(PG)method.The great performance of the fully SBGD is verified by experiments.This work will promote innovative development of the soft bionic grasping robots,and greatly meet the applications of dexterous grasping multi-size and multi-shape objects.