Feature Pyramid Network (FPN) has been an essential module for object detection models to consider various scales of an object. However, average precision (AP) on small objects is relatively lower than AP on medium and large objects. The reason is why the deeper layer of CNN causes information loss as feature extraction level. We propose a new scale sequence (S^2) feature extraction of FPN to strengthen feature information of small objects. We consider FPN structure as scale-space and extract scale sequence (S^2) feature by 3D convolution on the level axis of FPN. It is basically scale invariant feature and is built on high-resolution pyramid feature map for small objects. Furthermore, the proposed S^2 feature can be extended to most object detection models based on FPN. We demonstrate the proposed S2 feature can improve the performance of both one-stage and two-stage detectors on MS COCO dataset. Based on the proposed S2 feature, we achieve upto 1.3% and 1.1% of AP improvement for YOLOv4-P5 and YOLOv4-P6, respectively. For Faster RCNN and Mask R-CNN, we observe upto 2.0% and 1.6% of AP improvement with the suggested S^2 feature, respectively.
Data augmentation greatly increases the amount of data obtained based on labeled data to save on expenses and labor for data collection and labeling. We present a new approach for data augmentation called nine-dot MLS (ND-MLS). This approach is proposed based on the idea of image defor-mation. Images are deformed based on control points, which are calculated by ND-MLS. The method can generate over 2000 images for one exist-ing dataset in a short time. To verify this data augmentation method, extensive tests were performed covering 3 main tasks of computer vision, namely, classification, detection and segmentation. The results show that 1) in classification, 10 images per category were used for training, and VGGNet can obtain 92% top-1 acc on the MNIST dataset of handwritten digits by ND-MLS. In the Omniglot dataset, the few-shot accuracy usu-ally decreases with the increase in character categories. However, the ND-MLS method has stable performance and obtains 96.5 top-1 acc in Res-Net on 100 different handwritten character classification tasks; 2) in segmentation, under the premise of only ten original images, DeepLab obtains 93.5%, 85%, and 73.3% m_IOU(10) on the bottle, horse, and grass test datasets, respectively, while the cat test dataset obtains 86.7% m_IOU(10) with the SegNet model; 3) with only 10 original images from each category in object detection, YOLO v4 obtains 100% and 97.2% bottle and horse detection, respectively, while the cat dataset obtains 93.6% with YOLO v3. In summary, ND-MLS can perform well on classification, object detec-tion, and semantic segmentation tasks by using only a few data.
Over the last decade, multi-tasking learning approaches have achieved promising results in solving panoptic driving perception problems, providing both high-precision and high-efficiency performance. It has become a popular paradigm when designing networks for real-time practical autonomous driving system, where computation resources are limited. This paper proposed an effective and efficient multi-task learning network to simultaneously perform the task of traffic object detection, drivable road area segmentation and lane detection. Our model achieved the new state-of-the-art (SOTA) performance in terms of accuracy and speed on the challenging BDD100K dataset. Especially, the inference time is reduced by half compared to the previous SOTA model. Code will be released in the near future.
Multi-scale features have been proven highly effective for object detection but often come with huge and even prohibitive extra computation costs, especially for the recent Transformer-based detectors. In this paper, we propose Iterative Multi-scale Feature Aggregation (IMFA) -- a generic paradigm that enables efficient use of multi-scale features in Transformer-based object detectors. The core idea is to exploit sparse multi-scale features from just a few crucial locations, and it is achieved with two novel designs. First, IMFA rearranges the Transformer encoder-decoder pipeline so that the encoded features can be iteratively updated based on the detection predictions. Second, IMFA sparsely samples scale-adaptive features for refined detection from just a few keypoint locations under the guidance of prior detection predictions. As a result, the sampled multi-scale features are sparse yet still highly beneficial for object detection. Extensive experiments show that the proposed IMFA boosts the performance of multiple Transformer-based object detectors significantly yet with only slight computational overhead.
Existing top-performance 3D object detectors typically rely on the multi-modal fusion strategy. This design is however fundamentally restricted due to overlooking the modality-specific useful information and finally hampering the model performance. To address this limitation, in this work we introduce a novel modality interaction strategy where individual per-modality representations are learned and maintained throughout for enabling their unique characteristics to be exploited during object detection. To realize this proposed strategy, we design a DeepInteraction architecture characterized by a multi-modal representational interaction encoder and a multi-modal predictive interaction decoder. Experiments on the large-scale nuScenes dataset show that our proposed method surpasses all prior arts often by a large margin. Crucially, our method is ranked at the first position at the highly competitive nuScenes object detection leaderboard.
Current state-of-the-art object proposal networks are trained with a closed-world assumption, meaning they learn to only detect objects of the training classes. These models fail to provide high recall in open-world environments where important novel objects may be encountered. While a handful of recent works attempt to tackle this problem, they fail to consider that the optimal behavior of a proposal network can vary significantly depending on the data and application. Our goal is to provide a flexible proposal solution that can be easily tuned to suit a variety of open-world settings. To this end, we design a Tunable Hybrid Proposal Network (THPN) that leverages an adjustable hybrid architecture, a novel self-training procedure, and dynamic loss components to optimize the tradeoff between known and unknown object detection performance. To thoroughly evaluate our method, we devise several new challenges which invoke varying degrees of label bias by altering known class diversity and label count. We find that in every task, THPN easily outperforms existing baselines (e.g., RPN, OLN). Our method is also highly data efficient, surpassing baseline recall with a fraction of the labeled data.
The development of safety-oriented research and applications requires fine-grain vehicle trajectories that not only have high accuracy, but also capture substantial safety-critical events. However, it would be challenging to satisfy both these requirements using the available vehicle trajectory datasets do not have the capacity to satisfy both.This paper introduces the CitySim dataset that has the core objective of facilitating safety-oriented research and applications. CitySim has vehicle trajectories extracted from 1140 minutes of drone videos recorded at 12 locations. It covers a variety of road geometries including freeway basic segments, signalized intersections, stop-controlled intersections, and control-free intersections. CitySim was generated through a five-step procedure that ensured trajectory accuracy. The five-step procedure included video stabilization, object filtering, multi-video stitching, object detection and tracking, and enhanced error filtering. Furthermore, CitySim provides the rotated bounding box information of a vehicle, which was demonstrated to improve safety evaluations. Compared with other video-based critical events, including cut-in, merge, and diverge events, which were validated by distributions of both minimum time-to-collision and minimum post-encroachment time. In addition, CitySim had the capability to facilitate digital-twin-related research by providing relevant assets, such as the recording locations' three-dimensional base maps and signal timings.
3D Multi-Object Tracking (MOT) has achieved tremendous achievement thanks to the rapid development of 3D object detection and 2D MOT. Recent advanced works generally employ a series of object attributes, e.g., position, size, velocity, and appearance, to provide the clues for the association in 3D MOT. However, these cues may not be reliable due to some visual noise, such as occlusion and blur, leading to tracking performance bottleneck. To reveal the dilemma, we conduct extensive empirical analysis to expose the key bottleneck of each clue and how they correlate with each other. The analysis results motivate us to efficiently absorb the merits among all cues, and adaptively produce an optimal tacking manner. Specifically, we present Location and Velocity Quality Learning, which efficiently guides the network to estimate the quality of predicted object attributes. Based on these quality estimations, we propose a quality-aware object association (QOA) strategy to leverage the quality score as an important reference factor for achieving robust association. Despite its simplicity, extensive experiments indicate that the proposed strategy significantly boosts tracking performance by 2.2% AMOTA and our method outperforms all existing state-of-the-art works on nuScenes by a large margin. Moreover, QTrack achieves 48.0% and 51.1% AMOTA tracking performance on the nuScenes validation and test sets, which significantly reduces the performance gap between pure camera and LiDAR based trackers.
Detecting Out-of-Distribution (OOD) samples in real world visual applications like classification or object detection has become a necessary precondition in today's deployment of Deep Learning systems. Many techniques have been proposed, of which Energy based OOD methods have proved to be promising and achieved impressive performance. We propose semantic driven energy based method, which is an end-to-end trainable system and easy to optimize. We distinguish in-distribution samples from out-distribution samples with an energy score coupled with a representation score. We achieve it by minimizing the energy for in-distribution samples and simultaneously learn respective class representations that are closer and maximizing energy for out-distribution samples and pushing their representation further out from known class representation. Moreover, we propose a novel loss function which we call Cluster Focal Loss(CFL) that proved to be simple yet very effective in learning better class wise cluster center representations. We find that, our novel approach enhances outlier detection and achieve state-of-the-art as an energy-based model on common benchmarks. On CIFAR-10 and CIFAR-100 trained WideResNet, our model significantly reduces the relative average False Positive Rate(at True Positive Rate of 95%) by 67.2% and 57.4% respectively, compared to the existing energy based approaches. Further, we extend our framework for object detection and achieve improved performance.
In this work, we propose a novel uncertainty-aware object detection framework with a structured-graph, where nodes and edges are denoted by objects and their spatial-semantic similarities, respectively. Specifically, we aim to consider relationships among objects for effectively contextualizing them. To achieve this, we first detect objects and then measure their semantic and spatial distances to construct an object graph, which is then represented by a graph neural network (GNN) for refining visual CNN features for objects. However, refining CNN features and detection results of every object are inefficient and may not be necessary, as that include correct predictions with low uncertainties. Therefore, we propose to handle uncertain objects by not only transferring the representation from certain objects (sources) to uncertain objects (targets) over the directed graph, but also improving CNN features only on objects regarded as uncertain with their representational outputs from the GNN. Furthermore, we calculate a training loss by giving larger weights on uncertain objects, to concentrate on improving uncertain object predictions while maintaining high performances on certain objects. We refer to our model as Uncertainty-Aware Graph network for object DETection (UAGDet). We then experimentally validate ours on the challenging large-scale aerial image dataset, namely DOTA, that consists of lots of objects with small to large sizes in an image, on which ours improves the performance of the existing object detection network.
Taking into account information across the temporal domain helps to improve environment perception in autonomous driving. However, it has not been studied so far whether temporally fused neural networks are vulnerable to deliberately generated perturbations, i.e. adversarial attacks, or whether temporal history is an inherent defense against them. In this work, we study whether temporal feature networks for object detection are vulnerable to universal adversarial attacks. We evaluate attacks of two types: imperceptible noise for the whole image and locally-bound adversarial patch. In both cases, perturbations are generated in a white-box manner using PGD. Our experiments confirm, that attacking even a portion of a temporal input suffices to fool the network. We visually assess generated perturbations to gain insights into the functioning of attacks. To enhance the robustness, we apply adversarial training using 5-PGD. Our experiments on KITTI and nuScenes datasets demonstrate, that a model robustified via K-PGD is able to withstand the studied attacks while keeping the mAP-based performance comparable to that of an unattacked model.
Deep neural network based object detectors are continuously evolving and are used in a multitude of applications, each having its own set of requirements. While safety-critical applications need high accuracy and reliability, low-latency tasks need resource and energy-efficient networks. Real-time detectors, which are a necessity in high-impact real-world applications, are continuously proposed, but they overemphasize the improvements in accuracy and speed while other capabilities such as versatility, robustness, resource and energy efficiency are omitted. A reference benchmark for existing networks does not exist, nor does a standard evaluation guideline for designing new networks, which results in ambiguous and inconsistent comparisons. We, thus, conduct a comprehensive study on multiple real-time detectors (anchor-, keypoint-, and transformer-based) on a wide range of datasets and report results on an extensive set of metrics. We also study the impact of variables such as image size, anchor dimensions, confidence thresholds, and architecture layers on the overall performance. We analyze the robustness of detection networks against distribution shifts, natural corruptions, and adversarial attacks. Also, we provide a calibration analysis to gauge the reliability of the predictions. Finally, to highlight the real-world impact, we conduct two unique case studies, on autonomous driving and healthcare applications. To further gauge the capability of networks in critical real-time applications, we report the performance after deploying the detection networks on edge devices. Our extensive empirical study can act as a guideline for the industrial community to make an informed choice on the existing networks. We also hope to inspire the research community towards a new direction in the design and evaluation of networks that focuses on a bigger and holistic overview for a far-reaching impact.
A big convergence of language, vision, and multimodal pretraining is emerging. In this work, we introduce a general-purpose multimodal foundation model BEiT-3, which achieves state-of-the-art transfer performance on both vision and vision-language tasks. Specifically, we advance the big convergence from three aspects: backbone architecture, pretraining task, and model scaling up. We introduce Multiway Transformers for general-purpose modeling, where the modular architecture enables both deep fusion and modality-specific encoding. Based on the shared backbone, we perform masked "language" modeling on images (Imglish), texts (English), and image-text pairs ("parallel sentences") in a unified manner. Experimental results show that BEiT-3 obtains state-of-the-art performance on object detection (COCO), semantic segmentation (ADE20K), image classification (ImageNet), visual reasoning (NLVR2), visual question answering (VQAv2), image captioning (COCO), and cross-modal retrieval (Flickr30K, COCO).
Learning accurate depth is essential to multi-view 3D object detection. Recent approaches mainly learn depth from monocular images, which confront inherent difficulties due to the ill-posed nature of monocular depth learning. Instead of using a sole monocular depth method, in this work, we propose a novel Surround-view Temporal Stereo (STS) technique that leverages the geometry correspondence between frames across time to facilitate accurate depth learning. Specifically, we regard the field of views from all cameras around the ego vehicle as a unified view, namely surroundview, and conduct temporal stereo matching on it. The resulting geometrical correspondence between different frames from STS is utilized and combined with the monocular depth to yield final depth prediction. Comprehensive experiments on nuScenes show that STS greatly boosts 3D detection ability, notably for medium and long distance objects. On BEVDepth with ResNet-50 backbone, STS improves mAP and NDS by 2.6% and 1.4%, respectively. Consistent improvements are observed when using a larger backbone and a larger image resolution, demonstrating its effectiveness
Recent research in multi-task learning reveals the benefit of solving related problems in a single neural network. 3D object detection and multi-object tracking (MOT) are two heavily intertwined problems predicting and associating an object instance location across time. However, most previous works in 3D MOT treat the detector as a preceding separated pipeline, disjointly taking the output of the detector as an input to the tracker. In this work, we present Minkowski Tracker, a sparse spatio-temporal R-CNN that jointly solves object detection and tracking. Inspired by region-based CNN (R-CNN), we propose to solve tracking as a second stage of the object detector R-CNN that predicts assignment probability to tracks. First, Minkowski Tracker takes 4D point clouds as input to generate a spatio-temporal Bird's-eye-view (BEV) feature map through a 4D sparse convolutional encoder network. Then, our proposed TrackAlign aggregates the track region-of-interest (ROI) features from the BEV features. Finally, Minkowski Tracker updates the track and its confidence score based on the detection-to-track match probability predicted from the ROI features. We show in large-scale experiments that the overall performance gain of our method is due to four factors: 1. The temporal reasoning of the 4D encoder improves the detection performance 2. The multi-task learning of object detection and MOT jointly enhances each other 3. The detection-to-track match score learns implicit motion model to enhance track assignment 4. The detection-to-track match score improves the quality of the track confidence score. As a result, Minkowski Tracker achieved the state-of-the-art performance on Nuscenes dataset tracking task without hand-designed motion models.
3D object detection with surrounding cameras has been a promising direction for autonomous driving. In this paper, we present SimMOD, a Simple baseline for Multi-camera Object Detection, to solve the problem. To incorporate multi-view information as well as build upon previous efforts on monocular 3D object detection, the framework is built on sample-wise object proposals and designed to work in a two-stage manner. First, we extract multi-scale features and generate the perspective object proposals on each monocular image. Second, the multi-view proposals are aggregated and then iteratively refined with multi-view and multi-scale visual features in the DETR3D-style. The refined proposals are end-to-end decoded into the detection results. To further boost the performance, we incorporate the auxiliary branches alongside the proposal generation to enhance the feature learning. Also, we design the methods of target filtering and teacher forcing to promote the consistency of two-stage training. We conduct extensive experiments on the 3D object detection benchmark of nuScenes to demonstrate the effectiveness of SimMOD and achieve new state-of-the-art performance. Code will be available at https://github.com/zhangyp15/SimMOD.
Video object detection (VID) is challenging because of the high variation of object appearance as well as the diverse deterioration in some frames. On the positive side, the detection in a certain frame of a video, compared with that in a still image, can draw support from other frames. Hence, how to aggregate features across different frames is pivotal to VID problem. Most of existing aggregation algorithms are customized for two-stage detectors. However, these detectors are usually computationally expensive due to their two-stage nature. This work proposes a simple yet effective strategy to address the above concerns, which costs marginal overheads with significant gains in accuracy. Concretely, different from traditional two-stage pipeline, we select important regions after the one-stage detection to avoid processing massive low-quality candidates. Besides, we evaluate the relationship between a target frame and reference frames to guide the aggregation. We conduct extensive experiments and ablation studies to verify the efficacy of our design, and reveal its superiority over other state-of-the-art VID approaches in both effectiveness and efficiency. Our YOLOX-based model can achieve promising performance (\emph{e.g.}, 87.5\% AP50 at over 30 FPS on the ImageNet VID dataset on a single 2080Ti GPU), making it attractive for large-scale or real-time applications. The implementation is simple, we have made the demo codes and models available at \url{https://github.com/YuHengsss/YOLOV}.
We propose a new setting that relaxes an assumption in the conventional Co-Salient Object Detection (CoSOD) setting by allowing the presence of "noisy images" which do not show the shared co-salient object. We call this new setting Generalised Co-Salient Object Detection (GCoSOD). We propose a novel random sampling based Generalised CoSOD Training (GCT) strategy to distill the awareness of inter-image absence of co-salient objects into CoSOD models. It employs a Diverse Sampling Self-Supervised Learning (DS3L) that, in addition to the provided supervised co-salient label, introduces additional self-supervised labels for noisy images (being null, that no co-salient object is present). Further, the random sampling process inherent in GCT enables the generation of a high-quality uncertainty map highlighting potential false-positive predictions at instance level. To evaluate the performance of CoSOD models under the GCoSOD setting, we propose two new testing datasets, namely CoCA-Common and CoCA-Zero, where a common salient object is partially present in the former and completely absent in the latter. Extensive experiments demonstrate that our proposed method significantly improves the performance of CoSOD models in terms of the performance under the GCoSOD setting as well as the model calibration degrees.
Monocular 3D object detection is a fundamental but very important task to many applications including autonomous driving, robotic grasping and augmented reality. Existing leading methods tend to estimate the depth of the input image first, and detect the 3D object based on point cloud. This routine suffers from the inherent gap between depth estimation and object detection. Besides, the prediction error accumulation would also affect the performance. In this paper, a novel method named MonoSIM is proposed. The insight behind introducing MonoSIM is that we propose to simulate the feature learning behaviors of a point cloud based detector for monocular detector during the training period. Hence, during inference period, the learned features and prediction would be similar to the point cloud based detector as possible. To achieve it, we propose one scene-level simulation module, one RoI-level simulation module and one response-level simulation module, which are progressively used for the detector's full feature learning and prediction pipeline. We apply our method to the famous M3D-RPN detector and CaDDN detector, conducting extensive experiments on KITTI and Waymo Open datasets. Results show that our method consistently improves the performance of different monocular detectors for a large margin without changing their network architectures. Our codes will be publicly available at https://github.com/sunh18/MonoSIM}{https://github.com/sunh18/MonoSIM.
Currently, detecting 3D objects in Bird's-Eye-View (BEV) is superior to other 3D detectors for autonomous driving and robotics. However, transforming image features into BEV necessitates special operators to conduct feature sampling. These operators are not supported on many edge devices, bringing extra obstacles when deploying detectors. To address this problem, we revisit the generation of BEV representation and propose detecting objects in perspective BEV -- a new BEV representation that does not require feature sampling. We demonstrate that perspective BEV features can likewise enjoy the benefits of the BEV paradigm. Moreover, the perspective BEV improves detection performance by addressing issues caused by feature sampling. We propose PersDet for high-performance object detection in perspective BEV space based on this discovery. While implementing a simple and memory-efficient structure, PersDet outperforms existing state-of-the-art monocular methods on the nuScenes benchmark, reaching 34.6% mAP and 40.8% NDS when using ResNet-50 as the backbone.