General Detector Research Articles

Fine-Grained Prototypes Distillation for Few-Shot Object Detection

Few-shot object detection (FSOD) aims at extending a generic detector for novel object detection with only a few training examples. It attracts great concerns recently due to the practical meanings. Meta-learning has been demonstrated to be an effective paradigm for this task. In general, methods based on meta-learning employ an additional support branch to encode novel examples (a.k.a. support images) into class prototypes, which are then fused with query branch to facilitate the model prediction. However, the class-level prototypes are difficult to precisely generate, and they also lack detailed information, leading to instability in performance. New methods are required to capture the distinctive local context for more robust novel object detection. To this end, we propose to distill the most representative support features into fine-grained prototypes. These prototypes are then assigned into query feature maps based on the matching results, modeling the detailed feature relations between two branches. This process is realized by our Fine-Grained Feature Aggregation (FFA) module. Moreover, in terms of high-level feature fusion, we propose Balanced Class-Agnostic Sampling (B-CAS) strategy and Non-Linear Fusion (NLF) module from differenct perspectives. They are complementary to each other and depict the high-level feature relations more effectively. Extensive experiments on PASCAL VOC and MS COCO benchmarks show that our method sets a new state-of-the-art performance in most settings. Our code is available at https://github.com/wangchen1801/FPD.

Potentiality of automatic parameter tuning suite available in ACTS track reconstruction software framework

Particle tracking is among the most sophisticated and complex part of the full event reconstruction chain. A number of reconstruction algorithms work in a sequence to build these trajectories from detector hits. Each of these algorithms use many configuration parameters that need to be fine-tuned to properly account for the detector/experimental setup, the available CPU budget and the desired physics performance. Few examples of such parameters include the cut values limiting the search space of the algorithm, the approximations accounting for complex phenomena or the parameters controlling algorithm performance. The most popular method to tune these parameters is hand-tuning using brute-force techniques. These techniques can be inefficient and raise issues for the long-term maintainability of such algorithms. The opensource track reconstruction software framework known as “A Common Tracking Framework (ACTS)” offers an alternative solution to these parameter tuning techniques through the use of automatic parameter optimization algorithms. ACTS come equipped with an auto-tuning suite that provides necessary setup for performing optimization of input parameters belonging to track reconstruction algorithms. The user can choose the tunable parameters in a flexible way and define a cost/benefit function for optimizing the full reconstruction chain. The fast execution speed of ACTS allows the user to run several iterations of optimization within a reasonable time bracket. The performance of these optimizers has been demonstrated on different track reconstruction algorithms such as trajectory seed reconstruction and selection, particle vertex reconstruction and generation of simplified material map, and on different detector geometries such as Generic Detector and Open Data Detector (ODD). We aim to bring this approach to all aspects of trajectory reconstruction by having a more flexible integration of tunable parameters within ACTS.

Toward reliable fusion object detection based on dilated pyramid and semantic attention

AbstractObject detection on fused images of visible and infrared modals is of great importance for many applications, for example, surveillance and rescue at low‐light conditions. However, current detectors have difficulty for robust fused image detection for mainly two reasons. First, objects are presented in various shapes and sizes, making some hard samples cannot be localized accurately. Second, the same object category in the fused images will have different appearance due to changing weather condition, temperature and intrinsic heat. Such a contradiction will degrade the classification task of a detection network, since it cannot merge commonalities and distinguish differences well. In this paper, we propose to reconstruct the detection pipeline of current detectors, and enhance the detection ability on difficult samples in fused images. Specifically, a Dilation Pyramid Network (DPN) is designed at the lateral connection to generate and aggregate features of various receptive field, without increasing pyramid layers. To strengthen the classification, a Semantic Category Attention Module (SCAM) is proposed to capture attention centers of semantics in fused images, rather than object centers. Abundant experiments on two fusion datasets show that the proposed method achieves a satisfying performance, and both modules can greatly improve current generic detectors on fused images.

Boosting R-CNN: Reweighting R-CNN samples by RPN’s error for underwater object detection

Complicated underwater environments bring new challenges to object detection, such as unbalanced light conditions, low contrast, occlusion, and mimicry of aquatic organisms. Under these circumstances, the objects captured by the underwater camera will become vague, and the generic detectors often fail on these vague objects. This work aims to solve the problem from two perspectives: uncertainty modeling and hard example mining. We propose a two-stage underwater detector named boosting R-CNN, which comprises three key components. First, a new region proposal network named RetinaRPN is proposed, which provides high-quality proposals and considers objectness and IoU prediction for uncertainty to model the object prior probability. Second, the probabilistic inference pipeline is introduced to combine the first-stage prior uncertainty and the second-stage classification score to model the final detection score. Finally, we propose a new hard example mining method named boosting reweighting. Specifically, when the region proposal network miscalculates the object prior probability for a sample, boosting reweighting will increase the classification loss of the sample in the R-CNN head during training, while reducing the loss of easy samples with accurately estimated priors. Thus, a robust detection head in the second stage can be obtained. During the inference stage, the R-CNN has the capability to rectify the error of the first stage to improve the performance. Comprehensive experiments on two underwater datasets and two generic object detection datasets demonstrate the effectiveness and robustness of our method. The link of code:https://github.com/mousecpn/Boosting-R-CNN.

Efficient Few-Shot Object Detection via Knowledge Inheritance.

Few-shot object detection (FSOD), which aims at learning a generic detector that can adapt to unseen tasks with scarce training samples, has witnessed consistent improvement recently. However, most existing methods ignore the efficiency issues, e.g., high computational complexity and slow adaptation speed. Notably, efficiency has become an increasingly important evaluation metric for few-shot techniques due to an emerging trend toward embedded AI. To this end, we present an efficient pretrain-transfer framework (PTF) baseline with no computational increment, which achieves comparable results with previous state-of-the-art (SOTA) methods. Upon this baseline, we devise an initializer named knowledge inheritance (KI) to reliably initialize the novel weights for the box classifier, which effectively facilitates the knowledge transfer process and boosts the adaptation speed. Within the KI initializer, we propose an adaptive length re-scaling (ALR) strategy to alleviate the vector length inconsistency between the predicted novel weights and the pretrained base weights. Finally, our approach not only achieves the SOTA results across three public benchmarks, i.e., PASCAL VOC, COCO and LVIS, but also exhibits high efficiency with 1.8-100× faster adaptation speed against the other methods on COCO/LVIS benchmark during few-shot transfer. To our best knowledge, this is the first work to consider the efficiency problem in FSOD. We hope to motivate a trend toward powerful yet efficient few-shot technique development. The codes are publicly available at https://github.com/Ze-Yang/Efficient-FSOD.

Category Knowledge-guided Parameter Calibration for Few-shot Object Detection.

Few-shot object detection (FSOD) aims to adapt generic detectors to the novel categories with only a few annotations, which is an important and realistic task. Although the generic object detection has been widely studied over the past years, the FSOD is under explored. In this paper, we propose a novel Category Knowledge-guided Parameter Calibration (CKPC) framework to solve the FSOD task. We first propagate the category relation information to explore the representative category knowledge. Then, we explore the RoI-RoI and RoI-Category relations to capture the local-global context information to enhance the RoI (Region of Interest) features. Next, we project the knowledge representations of foreground categories into a parameter space by a linear transformation to generate the parameters of the category-level classifier. For the background, we learn a proxy category by concluding the global characteristics of all foreground categories to help ensure the discrepancy between the foreground and background, which is then projected into the parameter space by the same linear transformation. Finally, we leverage the parameters of the category-level classifier to explicitly calibrate the instance-level classifier learned on the enhanced RoI features for both the foreground and background categories to improve the detection performance. We conduct extensive experiments on two popular FSOD benchmarks (i.e., Pascal VOC and MS COCO), and the experimental results show that the proposed framework can outperform state-of-the-art methods.

Detector requirements for model-independent measurements of ultrahigh energy neutrino cross sections

The ultrahigh energy range of neutrino physics (above $\ensuremath{\sim}{10}^{7}\text{ }\text{ }\mathrm{GeV}$), as yet devoid of detections, is an open landscape with challenges to be met and discoveries to be made. Neutrino-nucleon cross sections in that range---with center-of-momentum energies $\sqrt{s}\ensuremath{\gtrsim}4\text{ }\text{ }\mathrm{TeV}$---are powerful probes of unexplored phenomena. We present a simple and accurate model-independent framework to evaluate how well these cross sections can be measured for an unknown flux and generic detectors. We also demonstrate how to characterize and compare detector sensitivity. We show that cross sections can be measured to ${\ensuremath{\simeq}}_{\ensuremath{-}30}^{+65}%$ precision over $\sqrt{s}\ensuremath{\simeq}4--140\text{ }\text{ }\mathrm{TeV}$ (${E}_{\ensuremath{\nu}}={10}^{7}\ensuremath{-}{10}^{10}\text{ }\text{ }\mathrm{GeV}$) with modest energy and angular resolution and $\ensuremath{\simeq}10$ events per energy decade. Many allowed novel-physics models (extra dimensions, leptoquarks, etc.) produce much larger effects. In the distant future, with $\ensuremath{\simeq}100$ events at the highest energies, the precision would be $\ensuremath{\simeq}15%$, probing even QCD saturation effects.

Upgrading the beam telescopes at the DESY II Test Beam Facility

The DESY II Test Beam Facility is a key infrastructure for modern high energy physics detector development, providing particles with a small momentum spread in a range from 1 to 6GeV to user groups e.g. from the LHC experiments and Belle II as well as generic detector R&D. Beam telescopes are provided in all three test beam areas as precise tracking reference without time stamping, with triggered readout and a readout time of > 115 μs . If the highest available rates are used, multiple particles are traversing the telescopes within one readout frame, thus creating ambiguities that cannot be resolved without additional timing layers. Several upgrades are currently investigated and tested: Firstly, a fast monolithic pixel sensor, the TelePix, to provide precise track timing and triggering on a region of interest is proposed to overcome this limitation. The TelePix is a 180nm HV-CMOS sensor that has been developed jointly by DESY, KIT and the University of Heidelberg and designed at KIT. In this publication, the performance evaluation is presented: The difference between two amplifier designs is evaluated. A high hit detection efficiency of above 99.9% combined with a time resolution of below 4ns at negligible pixel noise rates is determined. Finally, the digital hit output to provide region of interest triggering is evaluated and shows a short absolute delay with respect to a traditional trigger scintillator as well as an excellent time resolution. Secondly, a fast LGAD plane has been proposed to provide a time resolution of a few 10 ps, which is foreseen to drastically improve the timing performance of the telescope. Time resolutions of below 70 ps have been determined in collaboration with the University of California, Santa Barbara.

Learning scene-specific object detectors based on a generative-discriminative model with minimal supervision

• We propose the GDM to partition the sample space for improving the discriminability of the scene-specific object detector. • We propose the online gradual optimization procedure to automatically improve the performance of the GDM-based detector. • We propose a self-learning framework to train the GDM-based detector by marking several bounding boxes in the first frame. One object class may show large variations due to diverse illuminations, backgrounds, and camera viewpoints in the multi-scene object detection task. Traditional object detection methods generally perform poorly under unconstrained video environments. To address this problem, many modern approaches provide deep hierarchical appearance representations for object detection. Most of these methods require time-consuming training procedures on large manually annotated sample sets. In this paper, we propose a self-learning object detection framework to resolve the multi-scene detection problem in a bottom-up manner. A scene-specific objector is obtained from an autonomous learning process triggered by marking several bounding boxes around an object in the first video frame via a mouse. Here, artificially labeled training data or generic detectors are not needed. This learning process is conveniently replicated many times in different surveillance scenarios and produces scene-specific detectors from various camera viewpoints. Obviously, the initial scene-specific detector, initialized by several bounding boxes, exhibits poor detection performance and is difficult to be improved by traditional online learning algorithms . Consequently, we propose the Generative-Discriminative model (GDM) based detection method to partition detection response space and assign each partition an individual descriptor that progressively achieves high classification accuracy . Online gradual optimization process is proposed to optimize the Generative-Discriminative model and focus on those hard samples lying near the decision boundary. Experimental results on nine video datasets show that our approach achieves comparable performance to that of robust supervised methods, and outperforms state-of-the-art scene-specific object detection methods under varying imaging conditions.

A Novel Detection Refinement Technique for Accurate Identification of Nephrops norvegicus Burrows in Underwater Imagery.

With the evolution of the convolutional neural network (CNN), object detection in the underwater environment has gained a lot of attention. However, due to the complex nature of the underwater environment, generic CNN-based object detectors still face challenges in underwater object detection. These challenges include image blurring, texture distortion, color shift, and scale variation, which result in low precision and recall rates. To tackle this challenge, we propose a detection refinement algorithm based on spatial–temporal analysis to improve the performance of generic detectors by suppressing the false positives and recovering the missed detections in underwater videos. In the proposed work, we use state-of-the-art deep neural networks such as Inception, ResNet50, and ResNet101 to automatically classify and detect the Norway lobster Nephrops norvegicus burrows from underwater videos. Nephrops is one of the most important commercial species in Northeast Atlantic waters, and it lives in burrow systems that it builds itself on muddy bottoms. To evaluate the performance of proposed framework, we collected the data from the Gulf of Cadiz. From experiment results, we demonstrate that the proposed framework effectively suppresses false positives and recovers missed detections obtained from generic detectors. The mean average precision (mAP) gained a 10% increase with the proposed refinement technique.

Enhanced Security Against Volumetric DDoS Attacks Using Adversarial Machine Learning

With the increasing number of Internet users, cybersecurity is becoming more and more critical. Denial of service (DoS) and distributed denial of service (DDoS) attacks are two of the most common types of attacks that can severely affect a website or a server and make them unavailable to other users. The number of DDoS attacks increased by 55% between the period January 2020 and March 2021. Some approaches for detecting the DoS and DDoS attacks employing different machine learning and deep learning techniques are reported in the literature. Recently, it is also observed that the attackers have started leveraging state-of-the-art AI tools such as generative models for generating synthetic attacks which fool the standard detectors. No concrete approach is reported for developing and training the models which are not only robust in the detection of standard DDoS attacks but which can also detect adversarial attacks which are created synthetically by the attackers with harmful intentions. To that end, in this work, we employ a generative adversarial network (GAN) to develop such a robust detector. The proposed framework can generate and classify the synthetic benign (normal) and malignant (DDoS) instances which are very similar to the corresponding real instances as evaluated by similarity scores. The GAN-based model also demonstrates how effectively the malicious actors can generate adversarial DDoS network traffic instances which look like normal instances using feature modification which are very difficult for the classifier to detect. An approach on how to make the classifiers robust enough to detect such kinds of deliberate adversarial attacks via modifying some specific attack features manually is also proposed. This work provides the first step towards developing a generic and robust detector for DDoS attacks originating from various sources.

Soundscape and ambient noise levels of the Arctic waters around Greenland

A longer Arctic open water season is expected to increase underwater noise levels due to anthropogenic activities such as shipping, seismic surveys, sonar, and construction. Many Arctic marine mammal species depend on sound for communication, navigation, and foraging, therefore quantifying underwater noise levels is critical for documenting change and providing input to management and legislation. Here we present long-term underwater sound recordings from 26 deployments around Greenland from 2011 to 2020. Ambient noise was analysed in third octave and decade bands and further investigated using generic detectors searching for tonal and transient sounds. Ambient noise levels partly overlap with previous Arctic observations, however we report much lower noise levels than previously documented, specifically for Melville Bay and the Greenland Sea. Consistent seasonal noise patterns occur in Melville Bay, Baffin Bay and Greenland Sea, with noise levels peaking in late summer and autumn correlating with open water periods and seismic surveys. These three regions also had similar tonal detection patterns that peaked in May/June, likely due to bearded seal vocalisations. Biological activity was more readily identified using detectors rather than band levels. We encourage additional research to quantify proportional noise contributions from geophysical, biological, and anthropogenic sources in Arctic waters.

Unsupervised domain-adaptive scene-specific pedestrian detection for static video surveillance

Objects from one category may be drawn from different distributions due to diverse illuminations, backgrounds, and camera viewpoints. Traditional object detection methods generally perform poorly due to the domain shift. To address this problem, we propose to train a domain-adaptive scene-specific pedestrian detector in an unsupervised manner. A generic detector is transferred to different target domains from one labeled source domain dataset without human-annotated target samples. Specifically, we first extend the generic detector to a dual-boundary classifier and collect hard samples as unlabeled target samples according to the detection confidence. Then, we propose a cycle semantic transfer network to align the instance-level and class-level distributions between the source domain and target domain and automatically label the hard samples. The initial generic detector is then re-trained by these labeled hard samples and specialized to a target scene. This process can be conveniently extended to different surveillance scenarios and generate specific detectors under various static camera viewpoints. Moreover, to reduce the impact of mislabeled hard samples on the generic detector, an online gradual optimization algorithm is proposed to iteratively update the generic model, thereby obtaining an optimized process that is insensitive to individual mislabeled target samples. Extensive experiments show that even if the target domain is not manually annotated, the proposed self-learning method demonstrates the effectiveness of pedestrian detection in various domain shift scenarios, and it outperforms existing scene-specific pedestrian detection methods and some classic supervised methods.

Dimuons in LHC

In a hadron collider like the Large Hadron Collider (LHC), the dimuon channel provides an ideal tool for the discovery of many predicted particles from Beyond Standard Model (BSM) theories such as high mass resonant, or non-resonant states. The high mass resonances like Z’ and the high mass non-resonant like contact in- teractions or extra dimensions should appear in the high mass tail of dimuon dis- tributions. The Drell-Yan process is the main source of high mass opposite sign muon pairs in the Standard Model (SM) which act as the irreducible background of signals of the above mentioned BSM models in proton-proton collisions. Therefore, it is important to study with the highest possible accuracy the Drell-Yan process. In this paper, we study in detail the expectations of the Drell-Yan process in the dimuon channel at next-to-next-to-leading order in QCD and next-to-leading order in the electroweak corrections. In our study, we considered a typical generic detector acceptance such as CMS or ATLAS at the LHC at center of mass energy 14 TeV. At the end of this study we gave estimates of the expected backgrounds for new physics searches in the dimuon final states.

TCM: Temporal Consistency Model for Head Detection in Complex Videos

Head detection in real-world videos is a classical research problem in computer vision. Head detection in videos is challenging than in a single image due to many nuisances that are commonly observed in natural videos, including arbitrary poses, appearances, and scales. Generally, head detection is treated as a particular case of object detection in a single image. However, the performance of object detectors deteriorates in unconstrained videos. In this paper, we propose a temporal consistency model (TCM) to enhance the performance of a generic object detector by integrating spatial-temporal information that exists among subsequent frames of a particular video. Generally, our model takes detection from a generic detector as input and improves mean average precision (mAP) by recovering missed detection and suppressing false positives. We compare and evaluate the proposed framework on four challenging datasets, i.e., HollywoodHeads, Casablanca, BOSS, and PAMELA. Experimental evaluation shows that the performance is improved by employing the proposed TCM model. We demonstrate both qualitatively and quantitatively that our proposed framework obtains significant improvements over other methods.

DD4hep a community driven detector description for HEP

Detector description is an essential component in simulation, reconstruction and analysis of data resulting from particle collisions in high energy physics experiments and for the detector development studies for future experiments. Current detector description implementations of running experiments are mostly specific implementations. DD4hep [1] is an open source toolkit created in 2012 to serve as a generic detector description solution. The main motivation behind DD4hep is to provide the community with an integrated solution for all these stages and address detector description in a broad sense, including the geometry and the materials used in the device, and additional parameters describing e.g. the detection techniques, constants required for alignment and calibration, description of the readout structures and conditions data. In these proceedings, we will give an overview of the project and discuss recent developments in DD4hep as well as showcase adaptions of the framework by LHC and upcoming accelerator projects together with the road map of future developments.

CMS Experience with Adoption of the Community supported DD4hep Toolkit

DD4hep is an open-source software toolkit that provides comprehensive and complete generic detector descriptions for high energy physics (HEP) detectors. The Compact Muon Solenoid collaboration (CMS) has recently evaluated and adopted DD4hep to replace its custom detector description software. CMS has demanding software requirements as a very large, longrunning experiment that must support legacy geometries and study many possible upgraded detector designs of a constantly evolving detector that will be taking data for many years to come. CMS has chosen DD4hep since it is a high-quality, community-supported solution that will benefit from continuing modernization and maintenance. This presentation will discuss the issues of DD4hep adoption, the advantages and disadvantages of the various design choices, performance results, and the integration of the plugin systems from CMS and Gaudi, another open-source software framework. Recommendations about DD4hep based upon the CMS use cases will also be presented.

Hierarchy independent sensitivity to leptonic δCP with atmospheric neutrinos

The Dirac leptonic CP phase {\delta}_CP is one of the crucial unknown parameters in neutrino oscillation physics. In this paper we explore the possibility of using low energy atmospheric neutrino events to probe {\delta}_CP . We show that at sub GeV energies, when the events are binned as a function of the energy and direction of the final state leptons, a consistent distinction between various true {\delta}_CP values is obtained. We also show that at these energies there is no sensitivity to the mass ordering/hierarchy, so that {\delta}_CP can be measured without hierarchy ambiguity. In addition a preliminary \c{hi}^2 analysis of the sensitivity to {\delta}_CP using atmospheric neutrinos assuming a generic detector with perfect separation between charged current {\nu}_{\mu} , \bar{{\nu}}_{\mu} , {\nu}_e and \bar{{\nu}}_e events is given.

Robust, real-time generic detector based on a multi-feature probabilistic method.

Robust, real-time event detection from physiological signals acquired during long-term ambulatory monitoring still represents a major challenge for highly-artifacted signals. In this paper, we propose an original and generic multi-feature probabilistic detector (MFPD) and apply it to real-time QRS complex detection under noisy conditions. The MFPD method calculates a binary Bayesian probability for each derived feature and makes a centralized fusion, using the Kullback-Leibler divergence. The method is evaluated on two ECG databases: 1) the MIT-BIH arrhythmia database from Physionet containing clean ECG signals, 2) a benchmark noisy database created by adding noise recordings of the MIT-BIH noise stress test database, also from Physionet, to the MIT-BIH arrhythmia database. Results are compared with a well-known wavelet-based detector, and two recently published detectors: one based on spatiotemporal characteristic of the QRS complex and the second, as the MFDP, based on feature calculations from the University of New South Wales detector (UNSW). For both benchmark Physionet databases, the proposed MFPD method achieves the lowest standard deviation in sensitivity and positive predictivity (+P) despite its online algorithm architecture. While the statistics are comparable for low-to mildly artifactual ECG signals, the MFPD outperforms reference methods for artifacted ECG with low SNR levels reaching 87.48 ± 14.21% in sensitivity and 89.39 ± 14.67% in +P as compared to 88.30 ± 17.66% and 86.06 ± 19.67% respectively from UNSW, the best performing reference method. With demonstrations on the extensively studied QRS detection problem, we consider that the proposed generic structure of the multi-feature probabilistic detector should offer promising perspectives for long-term monitoring applications for highly-artifacted signals.

Adaptive Deep Convolutional Neural Networks for Scene-Specific Object Detection

A deep convolutional neural network (CNN) becomes a widely used tool for object detection. Many previous works have achieved excellent performance on object detection benchmarks. However, these works present generic detectors whose performance will drop rapidly when they are applied to a surveillance scene. In this paper, we propose an efficient method to construct a scene-specific regression model based on a generic CNN-based classifier. Our regression model is an adaptive deep CNN (ADCNN), which can predict object locations in the surveillance scene. First, we transfer the generic CNN-based classifier to the surveillance scene by selecting useful kernels. Second, we learn the context information of the surveillance scene in our regression model for accurate location prediction. Our main contributions are: 1) a transfer learning method that selects useful kernels in the generic CNN-based classifier; 2) a special architecture that can effectively learn the local and global context information in the surveillance scene; and 3) a new objective function to effectively train parameters in ADCNN. Compared with some state-of-the-art models, ADCNN achieves the best performance on three surveillance data sets for pedestrian detection and one surveillance data set for vehicle detection.