We propose a novel and pragmatic framework for traffic scene perception with roadside cameras. The proposed framework covers a full-stack of roadside perception pipeline for infrastructure-assisted autonomous driving, including object detection, object localization, object tracking, and multi-camera information fusion. Unlike previous vision-based perception frameworks rely upon depth offset or 3D annotation at training, we adopt a modular decoupling design and introduce a landmark-based 3D localization method, where the detection and localization can be well decoupled so that the model can be easily trained based on only 2D annotations. The proposed framework applies to either optical or thermal cameras with pinhole or fish-eye lenses. Our framework is deployed at a two-lane roundabout located at Ellsworth Rd. and State St., Ann Arbor, MI, USA, providing 7x24 real-time traffic flow monitoring and high-precision vehicle trajectory extraction. The whole system runs efficiently on a low-power edge computing device with all-component end-to-end delay of less than 20ms.
We propose a new paradigm to automatically generate training data with accurate labels at scale using the text-toimage synthesis frameworks (e.g., DALL-E, Stable Diffusion, etc.). The proposed approach decouples training data generation into foreground object mask generation and background (context) image generation. For foreground object mask generation, we use a simple textual template with object class name as input to DALL-E to generate a diverse set of foreground images. A foreground-background segmentation algorithm is then used to generate foreground object masks. Next, in order to generate context images, first a language description of the context is generated by applying an image captioning method on a small set of images representing the context. These language descriptions are then used to generate diverse sets of context images using the DALL-E framework. These are then composited with object masks generated in the first step to provide an augmented training set for a classifier. We demonstrate the advantages of our approach on four object detection datasets including on Pascal VOC and COCO object detection tasks. Furthermore, we also highlight the compositional nature of our data generation approach on out-of-distribution and zero-shot data generation scenarios.
This work introduces Guardian Angel, an Android App that assists visually impaired people to avoid danger in complex traffic environment. The system, consisting of object detection by pretrained YOLO model, distance estimation and moving direction estimation, provides information about surrounding vehicles and alarms users of potential danger without expensive special purpose device. With an experiment of 8 subjects, we corroborate that in terms of satisfaction score in pedestrian-crossing experiment with the assistance of our App using a smartphone is better than when without under 99% confidence level. The time needed to cross a road is shorter on average with the assistance of our system, however, not reaching significant difference by our experiment. The App has been released in Google Play Store, open to the public for free.
The existing state-of-the-art (SOTA) video salient object detection (VSOD) models have widely followed short-term methodology, which dynamically determines the balance between spatial and temporal saliency fusion by solely considering the current consecutive limited frames. However, the short-term methodology has one critical limitation, which conflicts with the real mechanism of our visual system -- a typical long-term methodology. As a result, failure cases keep showing up in the results of the current SOTA models, and the short-term methodology becomes the major technical bottleneck. To solve this problem, this paper proposes a novel VSOD approach, which performs VSOD in a complete long-term way. Our approach converts the sequential VSOD, a sequential task, to a data mining problem, i.e., decomposing the input video sequence to object proposals in advance and then mining salient object proposals as much as possible in an easy-to-hard way. Since all object proposals are simultaneously available, the proposed approach is a complete long-term approach, which can alleviate some difficulties rooted in conventional short-term approaches. In addition, we devised an online updating scheme that can grasp the most representative and trustworthy pattern profile of the salient objects, outputting framewise saliency maps with rich details and smoothing both spatially and temporally. The proposed approach outperforms almost all SOTA models on five widely used benchmark datasets.
This paper presents a novel deep neural network framework for RGB-D salient object detection by controlling the message passing between the RGB images and depth maps on the feature level and exploring the long-range semantic contexts and geometric information on both RGB and depth features to infer salient objects. To achieve this, we formulate a dynamic message propagation (DMP) module with the graph neural networks and deformable convolutions to dynamically learn the context information and to automatically predict filter weights and affinity matrices for message propagation control. We further embed this module into a Siamese-based network to process the RGB image and depth map respectively and design a multi-level feature fusion (MFF) module to explore the cross-level information between the refined RGB and depth features. Compared with 17 state-of-the-art methods on six benchmark datasets for RGB-D salient object detection, experimental results show that our method outperforms all the others, both quantitatively and visually.
With the recent development of Semi-Supervised Object Detection (SS-OD) techniques, object detectors can be improved by using a limited amount of labeled data and abundant unlabeled data. However, there are still two challenges that are not addressed: (1) there is no prior SS-OD work on anchor-free detectors, and (2) prior works are ineffective when pseudo-labeling bounding box regression. In this paper, we present Unbiased Teacher v2, which shows the generalization of SS-OD method to anchor-free detectors and also introduces Listen2Student mechanism for the unsupervised regression loss. Specifically, we first present a study examining the effectiveness of existing SS-OD methods on anchor-free detectors and find that they achieve much lower performance improvements under the semi-supervised setting. We also observe that box selection with centerness and the localization-based labeling used in anchor-free detectors cannot work well under the semi-supervised setting. On the other hand, our Listen2Student mechanism explicitly prevents misleading pseudo-labels in the training of bounding box regression; we specifically develop a novel pseudo-labeling selection mechanism based on the Teacher and Student's relative uncertainties. This idea contributes to favorable improvement in the regression branch in the semi-supervised setting. Our method, which works for both anchor-free and anchor-based methods, consistently performs favorably against the state-of-the-art methods in VOC, COCO-standard, and COCO-additional.
We propose global context vision transformer (GC ViT), a novel architecture that enhances parameter and compute utilization for computer vision. Our method leverages global context self-attention modules, joint with standard local self-attention, to effectively and efficiently model both long and short-range spatial interactions, without the need for expensive operations such as computing attention masks or shifting local windows. In addition, we address the lack of the inductive bias in ViTs, and propose to leverage a modified fused inverted residual blocks in our architecture. Our proposed GC ViT achieves state-of-the-art results across image classification, object detection and semantic segmentation tasks. On ImageNet-1K dataset for classification, the variants of GC ViT with 51M, 90M and 201M parameters achieve 84.3%, 85.0% and 85.7% Top-1 accuracy, respectively, at 224 image resolution and without any pre-training, hence surpassing comparably-sized prior art such as CNN-based ConvNeXt and ViT-based MaxViT and Swin Transformer by a large margin. Pre-trained GC ViT backbones in downstream tasks of object detection, instance segmentation, and semantic segmentation using MS COCO and ADE20K datasets outperform prior work consistently. Specifically, GC ViT with a 4-scale DINO detection head achieves a box AP of 58.3 on MS COCO dataset.
Autonomous driving, in recent years, has been receiving increasing attention for its potential to relieve drivers' burdens and improve the safety of driving. In modern autonomous driving pipelines, the perception system is an indispensable component, aiming to accurately estimate the status of surrounding environments and provide reliable observations for prediction and planning. 3D object detection, which intelligently predicts the locations, sizes, and categories of the critical 3D objects near an autonomous vehicle, is an important part of a perception system. This paper reviews the advances in 3D object detection for autonomous driving. First, we introduce the background of 3D object detection and discuss the challenges in this task. Second, we conduct a comprehensive survey of the progress in 3D object detection from the aspects of models and sensory inputs, including LiDAR-based, camera-based, and multi-modal detection approaches. We also provide an in-depth analysis of the potentials and challenges in each category of methods. Additionally, we systematically investigate the applications of 3D object detection in driving systems. Finally, we conduct a performance analysis of the 3D object detection approaches, and we further summarize the research trends over the years and prospect the future directions of this area.
Multi-point tracking is a challenging task that involves detecting points in the scene and tracking them across a sequence of frames. Computing detection-based measures like the F-measure on a frame-by-frame basis is not sufficient to assess the overall performance, as it does not interpret performance in the temporal domain. The main evaluation metric available comes from Multi-object tracking (MOT) methods to benchmark performance on datasets such as KITTI with the recently proposed higher order tracking accuracy (HOTA) metric, which is capable of providing a better description of the performance over metrics such as MOTA, DetA, and IDF1. While the HOTA metric takes into account temporal associations, it does not provide a tailored means to analyse the spatial associations of a dataset in a multi-camera setup. Moreover, there are differences in evaluating the detection task for points when compared to objects (point distances vs. bounding box overlap). Therefore in this work, we propose a multi-view higher order tracking metric (mvHOTA) to determine the accuracy of multi-point (multi-instance and multi-class) tracking methods, while taking into account temporal and spatial associations.mvHOTA can be interpreted as the geometric mean of detection, temporal, and spatial associations, thereby providing equal weighting to each of the factors. We demonstrate the use of this metric to evaluate the tracking performance on an endoscopic point detection dataset from a previously organised surgical data science challenge. Furthermore, we compare with other adjusted MOT metrics for this use-case, discuss the properties of mvHOTA, and show how the proposed multi-view Association and the Occlusion index (OI) facilitate analysis of methods with respect to handling of occlusions. The code is available at https://github.com/Cardio-AI/mvhota.
We propose Unified-IO, a model that performs a large variety of AI tasks spanning classical computer vision tasks, including pose estimation, object detection, depth estimation and image generation, vision-and-language tasks such as region captioning and referring expression, to natural language processing tasks such as question answering and paraphrasing. Developing a single unified model for such a large variety of tasks poses unique challenges due to the heterogeneous inputs and outputs pertaining to each task, including RGB images, per-pixel maps, binary masks, bounding boxes, and language. We achieve this unification by homogenizing every supported input and output into a sequence of discrete vocabulary tokens. This common representation across all tasks allows us to train a single transformer-based architecture, jointly on over 90 diverse datasets in the vision and language fields. Unified-IO is the first model capable of performing all 7 tasks on the GRIT benchmark and produces strong results across 16 diverse benchmarks like NYUv2-Depth, ImageNet, VQA2.0, OK-VQA, Swig, VizWizGround, BoolQ, and SciTail, with no task-specific fine-tuning. Code and demos for Unified-IO are available at: https://unified-io.allenai.org.
Semantic segmentation is the problem of assigning a class label to every pixel in an image, and is an important component of an autonomous vehicle vision stack for facilitating scene understanding and object detection. However, many of the top performing semantic segmentation models are extremely complex and cumbersome, and as such are not suited to deployment onboard autonomous vehicle platforms where computational resources are limited and low-latency operation is a vital requirement. In this survey, we take a thorough look at the works that aim to address this misalignment with more compact and efficient models capable of deployment on low-memory embedded systems while meeting the constraint of real-time inference. We discuss several of the most prominent works in the field, placing them within a taxonomy based on their major contributions, and finally we evaluate the inference speed of the discussed models under consistent hardware and software setups that represent a typical research environment with high-end GPU and a realistic deployed scenario using low-memory embedded GPU hardware. Our experimental results demonstrate that many works are capable of real-time performance on resource-constrained hardware, while illustrating the consistent trade-off between latency and accuracy.
Active Learning is a very common yet powerful framework for iteratively and adaptively sampling subsets of the unlabeled sets with a human in the loop with the goal of achieving labeling efficiency. Most real world datasets have imbalance either in classes and slices, and correspondingly, parts of the dataset are rare. As a result, there has been a lot of work in designing active learning approaches for mining these rare data instances. Most approaches assume access to a seed set of instances which contain these rare data instances. However, in the event of more extreme rareness, it is reasonable to assume that these rare data instances (either classes or slices) may not even be present in the seed labeled set, and a critical need for the active learning paradigm is to efficiently discover these rare data instances. In this work, we provide an active data discovery framework which can mine unknown data slices and classes efficiently using the submodular conditional gain and submodular conditional mutual information functions. We provide a general algorithmic framework which works in a number of scenarios including image classification and object detection and works with both rare classes and rare slices present in the unlabeled set. We show significant accuracy and labeling efficiency gains with our approach compared to existing state-of-the-art active learning approaches for actively discovering these rare classes and slices.
Most object detection frameworks use backbone architectures originally designed for image classification, conventionally with pre-trained parameters on ImageNet. However, image classification and object detection are essentially different tasks and there is no guarantee that the optimal backbone for classification is also optimal for object detection. Recent neural architecture search (NAS) research has demonstrated that automatically designing a backbone specifically for object detection helps improve the overall accuracy. In this paper, we introduce a neural architecture adaptation method that can optimize the given backbone for detection purposes, while still allowing the use of pre-trained parameters. We propose to adapt both the micro- and macro-architecture by searching for specific operations and the number of layers, in addition to the output channel dimensions of each block. It is important to find the optimal channel depth, as it greatly affects the feature representation capability and computation cost. We conduct experiments with our searched backbone for object detection and demonstrate that our backbone outperforms both manually designed and searched state-of-the-art backbones on the COCO dataset.
Object detection has made substantial progress in the last decade, due to the capability of convolution in extracting local context of objects. However, the scales of objects are diverse and current convolution can only process single-scale input. The capability of traditional convolution with a fixed receptive field in dealing with such a scale variance problem, is thus limited. Multi-scale feature representation has been proven to be an effective way to mitigate the scale variance problem. Recent researches mainly adopt partial connection with certain scales, or aggregate features from all scales and focus on the global information across the scales. However, the information across spatial and depth dimensions is ignored. Inspired by this, we propose the multi-scale convolution (MSConv) to handle this problem. Taking into consideration scale, spatial and depth information at the same time, MSConv is able to process multi-scale input more comprehensively. MSConv is effective and computationally efficient, with only a small increase of computational cost. For most of the single-stage object detectors, replacing the traditional convolutions with MSConvs in the detection head can bring more than 2.5\% improvement in AP (on COCO 2017 dataset), with only 3\% increase of FLOPs. MSConv is also flexible and effective for two-stage object detectors. When extended to the mainstream two-stage object detectors, MSConv can bring up to 3.0\% improvement in AP. Our best model under single-scale testing achieves 48.9\% AP on COCO 2017 \textit{test-dev} split, which surpasses many state-of-the-art methods.
Domain adversarial training has been ubiquitous for achieving invariant representations and is used widely for various domain adaptation tasks. In recent times, methods converging to smooth optima have shown improved generalization for supervised learning tasks like classification. In this work, we analyze the effect of smoothness enhancing formulations on domain adversarial training, the objective of which is a combination of task loss (eg. classification, regression, etc.) and adversarial terms. We find that converging to a smooth minima with respect to (w.r.t.) task loss stabilizes the adversarial training leading to better performance on target domain. In contrast to task loss, our analysis shows that converging to smooth minima w.r.t. adversarial loss leads to sub-optimal generalization on the target domain. Based on the analysis, we introduce the Smooth Domain Adversarial Training (SDAT) procedure, which effectively enhances the performance of existing domain adversarial methods for both classification and object detection tasks. Our analysis also provides insight into the extensive usage of SGD over Adam in the community for domain adversarial training.
Pyramidal networks are standard methods for multi-scale object detection. Current researches on feature pyramid networks usually adopt layer connections to collect features from certain levels of the feature hierarchy, and do not consider the significant differences among them. We propose a better architecture of feature pyramid networks, named selective multi-scale learning (SMSL), to address this issue. SMSL is efficient and general, which can be integrated in both single-stage and two-stage detectors to boost detection performance, with nearly no extra inference cost. RetinaNet combined with SMSL obtains 1.8\% improvement in AP (from 39.1\% to 40.9\%) on COCO dataset. When integrated with SMSL, two-stage detectors can get around 1.0\% improvement in AP.
Recent self-supervised learning (SSL) methods have shown impressive results in learning visual representations from unlabeled images. This paper aims to improve their performance further by utilizing the architectural advantages of the underlying neural network, as the current state-of-the-art visual pretext tasks for SSL do not enjoy the benefit, i.e., they are architecture-agnostic. In particular, we focus on Vision Transformers (ViTs), which have gained much attention recently as a better architectural choice, often outperforming convolutional networks for various visual tasks. The unique characteristic of ViT is that it takes a sequence of disjoint patches from an image and processes patch-level representations internally. Inspired by this, we design a simple yet effective visual pretext task, coined SelfPatch, for learning better patch-level representations. To be specific, we enforce invariance against each patch and its neighbors, i.e., each patch treats similar neighboring patches as positive samples. Consequently, training ViTs with SelfPatch learns more semantically meaningful relations among patches (without using human-annotated labels), which can be beneficial, in particular, to downstream tasks of a dense prediction type. Despite its simplicity, we demonstrate that it can significantly improve the performance of existing SSL methods for various visual tasks, including object detection and semantic segmentation. Specifically, SelfPatch significantly improves the recent self-supervised ViT, DINO, by achieving +1.3 AP on COCO object detection, +1.2 AP on COCO instance segmentation, and +2.9 mIoU on ADE20K semantic segmentation.
Unlike RGB cameras that use visible light bands (384$\sim$769 THz) and Lidars that use infrared bands (361$\sim$331 THz), Radars use relatively longer wavelength radio bands (77$\sim$81 GHz), resulting in robust measurements in adverse weathers. Unfortunately, existing Radar datasets only contain a relatively small number of samples compared to the existing camera and Lidar datasets. This may hinder the development of sophisticated data-driven deep learning techniques for Radar-based perception. Moreover, most of the existing Radar datasets only provide 3D Radar tensor (3DRT) data that contain power measurements along the Doppler, range, and azimuth dimensions. As there is no elevation information, it is challenging to estimate the 3D bounding box of an object from 3DRT. In this work, we introduce KAIST-Radar (K-Radar), a novel large-scale object detection dataset and benchmark that contains 35K frames of 4D Radar tensor (4DRT) data with power measurements along the Doppler, range, azimuth, and elevation dimensions, together with carefully annotated 3D bounding box labels of objects on the roads. K-Radar includes challenging driving conditions such as adverse weathers (fog, rain, and snow) on various road structures (urban, suburban roads, alleyways, and highways). In addition to the 4DRT, we provide auxiliary measurements from carefully calibrated high-resolution Lidars, surround stereo cameras, and RTK-GPS. We also provide 4DRT-based object detection baseline neural networks (baseline NNs) and show that the height information is crucial for 3D object detection. And by comparing the baseline NN with a similarly-structured Lidar-based neural network, we demonstrate that 4D Radar is a more robust sensor for adverse weather conditions. All codes are available at https://github.com/kaist-avelab/k-radar.
The 3D Average Precision (3D AP) relies on the intersection over union between predictions and ground truth objects. However, camera-only detectors have limited depth accuracy, which may cause otherwise reasonable predictions that suffer from such longitudinal localization errors to be treated as false positives. We therefore propose variants of the 3D AP metric to be more permissive with respect to depth estimation errors. Specifically, our novel longitudinal error tolerant metrics, LET-3D-AP and LET-3D-APL, allow longitudinal localization errors of the prediction boxes up to a given tolerance. To evaluate the proposed metrics, we also construct a new test set for the Waymo Open Dataset, tailored to camera-only 3D detection methods. Surprisingly, we find that state-of-the-art camera-based detectors can outperform popular LiDAR-based detectors with our new metrics past at 10% depth error tolerance, suggesting that existing camera-based detectors already have the potential to surpass LiDAR-based detectors in downstream applications. We believe the proposed metrics and the new benchmark dataset will facilitate advances in the field of camera-only 3D detection by providing more informative signals that can better indicate the system-level performance.
Vision-language (VL) pre-training has recently received considerable attention. However, most existing end-to-end pre-training approaches either only aim to tackle VL tasks such as image-text retrieval, visual question answering (VQA) and image captioning that test high-level understanding of images, or only target region-level understanding for tasks such as phrase grounding and object detection. We present FIBER (Fusion-In-the-Backbone-based transformER), a new VL model architecture that can seamlessly handle both these types of tasks. Instead of having dedicated transformer layers for fusion after the uni-modal backbones, FIBER pushes multimodal fusion deep into the model by inserting cross-attention into the image and text backbones, bringing gains in terms of memory and performance. In addition, unlike previous work that is either only pre-trained on image-text data or on fine-grained data with box-level annotations, we present a two-stage pre-training strategy that uses both these kinds of data efficiently: (i) coarse-grained pre-training based on image-text data; followed by (ii) fine-grained pre-training based on image-text-box data. We conduct comprehensive experiments on a wide range of VL tasks, ranging from VQA, image captioning, and retrieval, to phrase grounding, referring expression comprehension, and object detection. Using deep multimodal fusion coupled with the two-stage pre-training, FIBER provides consistent performance improvements over strong baselines across all tasks, often outperforming methods using magnitudes more data. Code is available at https://github.com/microsoft/FIBER.