skip to main content

Title: Multi-level 3D CNN for Learning Multi-scale Spatial Features
3D object recognition accuracy can be improved by learning the multi-scale spatial features from 3D spatial geometric representations of objects such as point clouds, 3D models, surfaces, and RGB-D data. Current deep learning approaches learn such features either using structured data representations (voxel grids and octrees) or from unstructured representations (graphs and point clouds). Learning features from such structured representations is limited by the restriction on resolution and tree depth while unstructured representations creates a challenge due to non-uniformity among data samples. In this paper, we propose an end-to-end multi-level learning approach on a multi-level voxel grid to overcome these drawbacks. To demonstrate the utility of the proposed multi-level learning, we use a multi-level voxel representation of 3D objects to perform object recognition. The multi-level voxel representation consists of a coarse voxel grid that contains volumetric information of the 3D object. In addition, each voxel in the coarse grid that contains a portion of the object boundary is subdivided into multiple fine-level voxel grids. The performance of our multi-level learning algorithm for object recognition is comparable to dense voxel representations while using significantly lower memory.
; ; ; ;
Award ID(s):
Publication Date:
Journal Name:
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops
Sponsoring Org:
National Science Foundation
More Like this
  1. 3D CT point clouds reconstructed from the original CT images are naturally represented in real-world coordinates. Compared with CT images, 3D CT point clouds contain invariant geometric features with irregular spatial distributions from multiple viewpoints. This paper rethinks pulmonary nodule detection in CT point cloud representations. We first extract the multi-view features from a sparse convolutional (SparseConv) encoder by rotating the point clouds with different angles in the world coordinate. Then, to simultaneously learn the discriminative and robust spatial features from various viewpoints, a nodule proposal optimization schema is proposed to obtain coarse nodule regions by aggregating consistent nodule proposals prediction from multi-view features. Last, the multi-level features and semantic segmentation features extracted from a SparseConv decoder are concatenated with multi-view features for final nodule region regression. Experiments on the benchmark dataset (LUNA16) demonstrate the feasibility of applying CT point clouds in lung nodule detection task. Furthermore, we observe that by combining multi-view predictions, the performance of the proposed framework is greatly improved compared to single-view, while the interior texture features of nodules from images are more suitable for detecting nodules in small sizes.
  2. We present MultiBodySync, a novel, end-to-end trainable multi-body motion segmentation and rigid registration framework for multiple input 3D point clouds. The two non-trivial challenges posed by this multi-scan multibody setting that we investigate are: (i) guaranteeing correspondence and segmentation consistency across multiple input point clouds capturing different spatial arrangements of bodies or body parts; and (ii) obtaining robust motion-based rigid body segmentation applicable to novel object categories. We propose an approach to address these issues that incorporates spectral synchronization into an iterative deep declarative network, so as to simultaneously recover consistent correspondences as well as motion segmentation. At the same time, by explicitly disentangling the correspondence and motion segmentation estimation modules, we achieve strong generalizability across different object categories. Our extensive evaluations demonstrate that our method is effective on various datasets ranging from rigid parts in articulated objects to individually moving objects in a 3D scene, be it single-view or full point clouds.
  3. Learning sensorimotor control policies from high-dimensional images crucially relies on the quality of the underlying visual representations. Prior works show that structured latent space such as visual keypoints often outperforms unstructured representations for robotic control. However, most of these representations, whether structured or unstructured are learned in a 2D space even though the control tasks are usually performed in a 3D environment. In this work, we propose a framework to learn such a 3D geometric structure directly from images in an end-to-end unsupervised manner. The input images are embedded into latent 3D keypoints via a differentiable encoder which is trained to optimize both a multi-view consistency loss and downstream task objective. These discovered 3D keypoints tend to meaningfully capture robot joints as well as object movements in a consistent manner across both time and 3D space. The proposed approach outperforms prior state-of-art methods across a variety of reinforcement learning benchmarks. Code and videos at
  4. We present multiresolution tree-structured networks to process point clouds for 3D shape understanding and generation tasks. Our network represents a 3D shape as a set of locality-preserving 1D ordered list of points at multiple resolutions. This allows efficient feed-forward processing through 1D convolutions, coarse-to-fine analysis through a multi-grid architecture, and it leads to faster convergence and small memory footprint during training. The proposed tree-structured encoders can be used to classify shapes and outperform existing point-based architectures on shape classification benchmarks, while tree-structured decoders can be used for generating point clouds directly and they outperform existing approaches for image-to-shape inference tasks learned using the ShapeNet dataset. Our model also allows unsupervised learning of point-cloud based shapes by using a variational autoencoder, leading to higher-quality generated shapes.
  5. Medical image analysis using deep learning has recently been prevalent, showing great performance for various downstream tasks including medical image segmentation and its sibling, volumetric image segmentation. Particularly, a typical volumetric segmentation network strongly relies on a voxel grid representation which treats volumetric data as a stack of individual voxel `slices', which allows learning to segment a voxel grid to be as straightforward as extending existing image-based segmentation networks to the 3D domain. However, using a voxel grid representation requires a large memory footprint, expensive test-time and limiting the scalability of the solutions. In this paper, we propose Point-Unet, a novel method that incorporates the eciency of deep learning with 3D point clouds into volumetric segmentation. Our key idea is to rst predict the regions of interest in the volume by learning an attentional probability map, which is then used for sampling the volume into a sparse point cloud that is subsequently segmented using a point-based neural network. We have conducted the experiments on the medical volumetric segmentation task with both a small-scale dataset Pancreas and large-scale datasets BraTS18, BraTS19, and BraTS20 challenges. A comprehensive benchmark on di erent metrics has shown that our context-aware Point-Unet robustly outperforms the SOTAmore »voxel-based networks at both accuracies, memory usage during training, and time consumption during testing.« less