Autonomous driving has attracted tremendous attention especially in the past few years. The key techniques for a self-driving car include solving tasks like 3D map construction, self-localization, parsing road and understanding objects, which enable vehicles to reason act. However, large scale data set training system evaluation is still bottleneck developing robust perception models. In this paper, we present ApolloScape dataset [1] its applications autonomous driving. Compared with...

Scene parsing aims to assign a class (semantic) label for each pixel in an image. It is comprehensive analysis of Given the rise autonomous driving, pixel-accurate environmental perception expected be key enabling technical piece. However, providing large scale dataset design and evaluation scene algorithms, particular outdoor scenes, has been difficult. The per-pixel labelling process prohibitively expensive, limiting existing ones. In this paper, we present large-scale open dataset,...

In the 2D/3D object detection task, Intersection-over-Union (IoU) has been widely employed as an evaluation metric to evaluate performance of different detectors in testing stage. However, during training stage, common distance loss (e.g, L_1 or L_2) is often adopted function minimize discrepancy between predicted and ground truth Bounding Box (Bbox). To eliminate gap testing, IoU introduced for 2D [1] [2]. Unfortunately, all these approaches only work axis-aligned Boxes, which cannot be...

Autonomous driving has attracted remarkable attention from both industry and academia. An important task is to estimate 3D properties (e.g. translation, rotation shape) of a moving or parked vehicle on the road. This task, while critical, still under-researched in computer vision community – partially owing lack large scale fully-annotated car database suitable for autonomous research. In this paper, we contribute first instance understanding ApolloCar3D. The dataset contains 5,277 images...

Accurate detection of obstacles in 3D is an essential task for autonomous driving and intelligent transportation. In this work, we propose a general multimodal fusion framework FusionPainting to fuse the 2D RGB image point clouds at semantic level boosting object task. Especially, consists three main modules: multi-modal segmentation module, adaptive attention-based detector. First, information obtained Lidar based on approaches. Then results from different sensors are adaptively fused...

Existing LiDAR-based 3D object detectors usually focus on the single-frame detection, while ignoring spatiotemporal information in consecutive point cloud frames. In this paper, we propose an end-to-end online video detector that operates sequences. The proposed model comprises a spatial feature encoding component and aggregation component. former component, novel Pillar Message Passing Network (PMPNet) is to encode each discrete frame. It adaptively collects for pillar node from its...

Currently, in Autonomous Driving (AD), most of the 3D object detection frameworks (either anchor- or anchor-free-based) consider as a Bounding Box (BBox) regression problem. However, this compact representation is not sufficient to explore all information objects. To tackle problem, we propose simple but practical framework jointly predict BBox and instance segmentation. For segmentation, Spatial Embeddings (SEs) strategy assemble foreground points into their corresponding centers. Base on...

Existing deep learning-based approaches for monocular 3D object detection in autonomous driving often model the as a rotated cuboid while object’s geometric shape has been ignored. In this work, we propose an approach incorporating shape-aware 2D/3D constraints into framework. Specifically, employ neural network to learn distinguished 2D keypoints image domain and regress their corresponding coordinates local coordinate first. Then are built by these correspondences each boost performance....

Despite the remarkable progresses made in deep learning based depth map super-resolution (DSR), how to tackle real-world degradation low-resolution (LR) maps remains a major challenge. Existing DSR model is generally trained and tested on synthetic dataset, which very different from what would get real sensor. In this paper, we argue that models under setting are restrictive not effective dealing with realworld tasks. We make two contributions tackling of sensors. First, propose classify...

Recovering an absolute metric scale from a monocular camera is challenging but highly desirable problem for camera-based systems. By using different kinds of cues, various approaches have been proposed estimation, such as height and object size. In this paper, first, we summarize estimation approaches. Then, propose robust divide-and-conquer method based on the ground plane by analyzing advantages disadvantages estimated scale, effective correction strategy has to reduce drift during visual...

A novel, to the best of our knowledge, color computational ghost imaging scheme is presented for reconstruction a object image, which greatly simplifies experimental setup and shortens acquisition time. Compared conventional schemes, it only adopts one digital light projector project speckles single-pixel detector receive intensity, instead utilizing three monochromatic paths separately synthesizing branch results. Severe noise distortion, are common in imaging, can be removed by utilization...

Annotating the LiDAR point cloud is crucial for deep learning-based 3D object detection tasks. Due to expensive labeling costs, data augmentation has been taken as a necessary module and plays an important role in training neural network. "Copy" "paste" (i.e., GT-Aug) most commonly used strategy, however, occlusion between objects not into consideration. To handle above limitation, we propose rendering-based frame-work LiDAR-Aug) enrich boost performance of LiDAR-based detectors. The...

Single image-based dehazing has achieved remarkable progress with the development of deep learning technologies. End-to-end neural networks have been proposed to learn a direct hazy-to-clear image translation recover clear structures and edges cues from hazy inputs. However, frequency domain information is explored insufficiently lots intermediate structure texture related current are ignored, which limits performances approaches. To handle these limitations mentioned above, wavelet spatial...

Single image dehazing is a challenging and illposed problem due to severe information degeneration of images captured in hazy conditions. Remarkable progresses have been achieved by deep-learning based methods, where residual learning commonly used separate the into clear haze components. However, nature low similarity between components neglected, while lack constraint contrastive peculiarity two always restricts performance these approaches. To deal with problems, we propose an end-to-end...

Due to the domain differences and unbalanced disparity distribution across multiple datasets, current stereo matching approaches are commonly limited a specific dataset generalize poorly others. Such shift issue is usually addressed by substantial adaptation on costly target-domain ground-truth data, which cannot be easily obtained in practical settings. In this paper, we propose dig into uncertainty estimation for robust matching. Specifically, balance distribution, employ pixel-level...

In Autonomous Driving (AD), detection and tracking of obstacles on the roads is a critical task. Deep-learning based methods using annotated LiDAR data have been most widely adopted approach for this. Unfortunately, annotating 3D point cloud very challenging, time- money-consuming this letter, we propose novel simulator that augments real with synthetic (e.g., vehicles, pedestrians, other movable objects). Unlike previous simulators entirely rely CG (Computer Graphics) models game engines,...

The success of supervised learning-based single image depth estimation methods critically depends on the availability large-scale dense per-pixel annotations, which requires both laborious and expensive annotation process. Therefore, self-supervised are much desirable, attract significant attention recently. However, maps predicted by existing tend to be blurry with many details lost. To overcome these limitations, we propose a novel framework, named MLDA-Net, obtain shaper boundaries richer...

Abstract With the aid of a electro-opto-thermal model two-dimensional (2-D) verticalcavity surface-emitting laser (VCSEL) array, thermal characteristics 4×4 VCSEL array is simulated. It shown that there higher temperature in central region device, which limits optical output performance. In order to enhance power, novel with non-uniform oxidation aperture designed. As result, peak junction decreased, and non-uniformity bias current density among cells are improved obviously. Furthermore,...