Kunfeng Wang

Huiming Yang, Wenzhuo Liu, Yicheng Qiao et al.

The sparse cross-modality detector offers more advantages than its counterpart, the Bird's-Eye-View (BEV) detector, particularly in terms of adaptability for downstream tasks and computational cost savings. However, existing sparse detectors overlook the quality of token representation, leaving it with a sub-optimal foreground quality and limited performance. In this paper, we identify that the geometric structure preserved and the class distribution are the key to improving the performance of the sparse detector, and propose a Sparse Selector (SS). The core module of SS is Ray-Aware Supervision (RAS), which preserves rich geometric information during the training stage, and Class-Balanced Supervision, which adaptively reweights the salience of class semantics, ensuring that tokens associated with small objects are retained during token sampling. Thereby, outperforming other sparse multi-modal detectors in the representation of tokens. Additionally, we design Ray Positional Encoding (Ray PE) to address the distribution differences between the LiDAR modality and the image. Finally, we integrate the aforementioned module into an end-to-end sparse multi-modality detector, dubbed CrossRay3D. Experiments show that, on the challenging nuScenes benchmark, CrossRay3D achieves state-of-the-art performance with 72.4 mAP and 74.7 NDS, while running 1.84 faster than other leading methods. Moreover, CrossRay3D demonstrates strong robustness even in scenarios where LiDAR or camera data are partially or entirely missing.

Xiaolong Guo, Xiaosong Lan, Kunfeng Wang et al.

Instance segmentation aims to locate targets in the image and segment each target area at pixel level, which is one of the most important tasks in computer vision. Mask R-CNN is a classic method of instance segmentation, but we find that its predicted masks are unclear and inaccurate near contours. To cope with this problem, we draw on the idea of contour matching based on distance transformation image and propose a novel loss function, called contour loss. Contour loss is designed to specifically optimize the contour parts of the predicted masks, thus can assure more accurate instance segmentation. In order to make the proposed contour loss to be jointly trained under modern neural network frameworks, we design a differentiable k-step distance transformation image calculation module, which can approximately compute truncated distance transformation images of the predicted mask and corresponding ground-truth mask online. The proposed contour loss can be integrated into existing instance segmentation methods such as Mask R-CNN, and combined with their original loss functions without modification of the inference network structures, thus has strong versatility. Experimental results on COCO show that contour loss is effective, which can further improve instance segmentation performances.

Yonglin Tian, Lichao Huang, Xuesong Li et al.

Varying density of point clouds increases the difficulty of 3D detection. In this paper, we present a context-aware dynamic network (CADNet) to capture the variance of density by considering both point context and semantic context. Point-level contexts are generated from original point clouds to enlarge the effective receptive filed. They are extracted around the voxelized pillars based on our extended voxelization method and processed with the context encoder in parallel with the pillar features. With a large perception range, we are able to capture the variance of features for potential objects and generate attentive spatial guidance to help adjust the strengths for different regions. In the region proposal network, considering the limited representation ability of traditional convolution where same kernels are shared among different samples and positions, we propose a decomposable dynamic convolutional layer to adapt to the variance of input features by learning from local semantic context. It adaptively generates the position-dependent coefficients for multiple fixed kernels and combines them to convolve with local feature windows. Based on our dynamic convolution, we design a dual-path convolution block to further improve the representation ability. We conduct experiments with our Network on KITTI dataset and achieve good performance on 3D detection task for both precision and speed. Our one-stage detector outperforms SECOND and PointPillars by a large margin and achieves the speed of 30 FPS.

Yonglin Tian, Kunfeng Wang, Yuang Wang et al.

This paper focuses on the construction of stronger local features and the effective fusion of image and LiDAR data. We adopt different modalities of LiDAR data to generate richer features and present an adaptive and azimuth-aware network to aggregate local features from image, bird's eye view maps and point cloud. Our network mainly consists of three subnetworks: ground plane estimation network, region proposal network and adaptive fusion network. The ground plane estimation network extracts features of point cloud and predicts the parameters of a plane which are used for generating abundant 3D anchors. The region proposal network generates features of image and bird's eye view maps to output region proposals. To integrate heterogeneous image and point cloud features, the adaptive fusion network explicitly adjusts the intensity of multiple local features and achieves the orientation consistency between image and LiDAR data by introduce an azimuth-aware fusion module. Experiments are conducted on KITTI dataset and the results validate the advantages of our aggregation of multimodal local features and the adaptive fusion network.

Kunfeng Wang, Chao Gou, Fei-Yue Wang

In this paper, we propose a robust change detection method for intelligent visual surveillance. This method, named M4CD, includes three major steps. Firstly, a sample-based background model that integrates color and texture cues is built and updated over time. Secondly, multiple heterogeneous features (including brightness variation, chromaticity variation, and texture variation) are extracted by comparing the input frame with the background model, and a multi-source learning strategy is designed to online estimate the probability distributions for both foreground and background. The three features are approximately conditionally independent, making multi-source learning feasible. Pixel-wise foreground posteriors are then estimated with Bayes rule. Finally, the Markov random field (MRF) optimization and heuristic post-processing techniques are used sequentially to improve accuracy. In particular, a two-layer MRF model is constructed to represent pixel-based and superpixel-based contextual constraints compactly. Experimental results on the CDnet dataset indicate that M4CD is robust under complex environments and ranks among the top methods.

Wenwen Zhang, Kunfeng Wang, Hua Qu et al.

As a special type of object detection, pedestrian detection in generic scenes has made a significant progress trained with large amounts of labeled training data manually. While the models trained with generic dataset work bad when they are directly used in specific scenes. With special viewpoints, flow light and backgrounds, datasets from specific scenes are much different from the datasets from generic scenes. In order to make the generic scene pedestrian detectors work well in specific scenes, the labeled data from specific scenes are needed to adapt the models to the specific scenes. While labeling the data manually spends much time and money, especially for specific scenes, each time with a new specific scene, large amounts of images must be labeled. What's more, the labeling information is not so accurate in the pixels manually and different people make different labeling information. In this paper, we propose an ACP-based method, with augmented reality's help, we build the virtual world of specific scenes, and make people walking in the virtual scenes where it is possible for them to appear to solve this problem of lacking labeled data and the results show that data from virtual world is helpful to adapt generic pedestrian detectors to specific scenes.

Yonglin Tian, Xuan Li, Kunfeng Wang et al.

In the area of computer vision, deep learning has produced a variety of state-of-the-art models that rely on massive labeled data. However, collecting and annotating images from the real world has a great demand for labor and money investments and is usually too passive to build datasets with specific characteristics, such as small area of objects and high occlusion level. Under the framework of Parallel Vision, this paper presents a purposeful way to design artificial scenes and automatically generate virtual images with precise annotations. A virtual dataset named ParallelEye is built, which can be used for several computer vision tasks. Then, by training the DPM (Deformable Parts Model) and Faster R-CNN detectors, we prove that the performance of models can be significantly improved by combining ParallelEye with publicly available real-world datasets during the training phase. In addition, we investigate the potential of testing the trained models from a specific aspect using intentionally designed virtual datasets, in order to discover the flaws of trained models. From the experimental results, we conclude that our virtual dataset is viable to train and test the object detectors.

Xuan Li, Kunfeng Wang, Yonglin Tian et al.

Video image datasets are playing an essential role in design and evaluation of traffic vision algorithms. Nevertheless, a longstanding inconvenience concerning image datasets is that manually collecting and annotating large-scale diversified datasets from real scenes is time-consuming and prone to error. For that virtual datasets have begun to function as a proxy of real datasets. In this paper, we propose to construct large-scale artificial scenes for traffic vision research and generate a new virtual dataset called "ParallelEye". First of all, the street map data is used to build 3D scene model of Zhongguancun Area, Beijing. Then, the computer graphics, virtual reality, and rule modeling technologies are utilized to synthesize large-scale, realistic virtual urban traffic scenes, in which the fidelity and geography match the real world well. Furthermore, the Unity3D platform is used to render the artificial scenes and generate accurate ground-truth labels, e.g., semantic/instance segmentation, object bounding box, object tracking, optical flow, and depth. The environmental conditions in artificial scenes can be controlled completely. As a result, we present a viable implementation pipeline for constructing large-scale artificial scenes for traffic vision research. The experimental results demonstrate that this pipeline is able to generate photorealistic virtual datasets with low modeling time and high accuracy labeling.