Search

The most recent 3D object detectors for point clouds rely on the coarse voxel-based representation rather than the accurate point-based representation due to a higher box recall in the voxel-based Region Proposal Network (RPN). However, the detection accuracy is severely restricted by the information loss of pose details in the voxels. Different from considering the point cloud as voxel or point representation only, we propose a point-to-voxel feature learning approach to voxelize the point cloud with both the point-wise semantic and local spatial features, which maintains the voxel-wise features to build the high-recall voxel-based RPN and also provides the accurate point-wise features for refining the detection results. Another difficulty in object detection for point cloud is that the visible part varies a lot against the full view of object because of the perspective issues in data acquisition. To address this, we propose an attentive corner aggregation module to attentively aggregate the features of local point cloud surrounding a 3D proposal from the perspectives of eight corners in the proposal 3D bounding box. The experimental results on the competitive KITTI 3D object detection benchmark show that the proposed method achieves state-of-the-art performance.

Motivated by the success of convolutional neural networks (CNNs) in image-related applications, in this paper, we design an effective method for no-reference 3D image quality assessment (3D IQA) through CNN-based feature extraction and consolidation strategy. In the first and most vital stage, quality-aware features, which reflect the inherent quality of images, are extracted by a fine-tuned CNN model exploiting the concept of transfer learning. This fine-tuning strategy solves the large-scale training data dependence existing in current deep-learning-based IQA algorithms. In the second stage, features from the left and right view are consolidated by linear weighted fusion where the weight for each image is obtained from its saliency map. In addition, the statistical characteristics of the disparity map are also considered in a multi-scale manner as additional features. In the final stage of quality mapping, the objective score for each stereoscopic pair is gained by support vector regression. The experimental results on the public databases show that our approach outperforms many existing no-reference and even full-reference methods.

The pervasion of 3-D technologies over the years gives rise to the increasing demands of accurate and efficient stereoscopic image quality assessment (SIQA) methods, designed to automatically supervise and optimize 3-D image and video processing systems. Though 2-D IQA has attracted considerable attention, its 3-D counterpart is yet to be well explored. In this paper, a no-reference SIQA method using convolution neural network (CNN) for feature extraction is proposed. In the proposed method, a CNN model is trained from scratch to classify images according to their perceptual quality, with quality-aware monocular features extracted from a higher level layer of the network. Then, visual saliency models are utilized to fuse the captured monocular features. In the meanwhile, multi-scale statistical features are derived from the binocular disparity maps. Finally, the fused CNN features and the disparity features are synthesized by support vector regression into the objective quality score of the stereoscopic image. Experimental results on two public databases demonstrate the superior performance of the proposed method over other state-of-the-art methods, in terms of its accuracy in predicting stereoscopic image quality as well as its robustness across different databases and distortion types.

With the booming of 3-D image processing in the entertainment industry and 3-D multimedia applications todays, the technology for assessing the quality of stereoscopic image faces more challenging tasks than its 2-D counterparts, such as binocular combination, stereo matching, and binocular rivalry. In this paper, a novel stereoscopic image quality assessment method is proposed by jointly exploring binocular fusion and rivalry models. In the quality-aware feature extraction stage, multi-scale binocular combination and binocular rivalry energy responses are first generated from reference and distorted stereopairs. In addition, considering that the changes of luminance affect the quality of stereoscopic image greatly, multiscale visual features relating to image quality are obtained from its luminance maps as another binocular combination and rivalry features. Then, the dissimilarity of quality-aware features between the reference stereopair and its distorted version is quantified. Finally, such dissimilarities are mapped into an objective score to represent the perceptual quality of stereoscopic image through the pooling strategy of support vector regression. Experiments conducted on LIVE 3-D databases demonstrate that the proposed method achieves 96.61% and 96.03% in terms of Pearson's linear correlation coefficient on Database Phase I and Phase II, respectively, outperforming most of the state-of-the-art methods.