Detection Test Set Research Articles

Large-Scale Object Detection in the Wild With Imbalanced Data Distribution, and Multi-Labels.

Training with more data has always been the most stable and effective way of improving performance in the deep learning era. The Open Images dataset, the largest object detection dataset, presents significant opportunities and challenges for general and sophisticated scenarios. However, its semi-automatic collection and labeling process, designed to manage the huge data scale, leads to label-related problems, including explicit or implicit multiple labels per object and highly imbalanced label distribution. In this work, we quantitatively analyze the major problems in large-scale object detection and provide a detailed yet comprehensive demonstration of our solutions. First, we design a concurrent softmax to handle the multi-label problems in object detection and propose a soft-balance sampling method with a hybrid training scheduler to address the label imbalance. This approach yields a notable improvement of 3.34 points, achieving the best single-model performance with a mAP of 60.90% on the public object detection test set of Open Images. Then, we introduce a well-designed ensemble mechanism that substantially enhances the performance of the single model, achieving an overall mAP of 67.17%, which is 4.29 points higher than the best result from the Open Images public test 2018. Our result is published on https://www.kaggle.com/c/open-images-2019-object-detection/leaderboard.

Diagnostic Test Point Insertion and Test Compaction

Test points are inserted into a circuit to improve its testability or diagnosability. The diagnosability goal may be to reduce the number of indistinguished fault pairs, increase the accuracy of logic diagnosis, or reduce the number of diagnostic tests. This article describes a test point insertion procedure, combined with a test compaction procedure, whose goal is to reduce the number of tests in a diagnostic test set, and bring it closer to the number of tests in a fault detection test set. A test point allows a fan-out branch to assume values that are different from those of the fan-out stem. Effectively, the fan-out branch becomes an independent input. The test compaction procedure reduces the number of tests in two ways: 1) by utilizing the extra inputs created when test points are inserted and 2) a diagnostic test set typically consists of a fault detection test set followed by diagnostic tests. The test compaction procedure modifies the diagnostic test set to use all the tests as both fault detection and diagnostic tests. Experimental results for single stuck-at faults in benchmark circuits demonstrate the effectiveness of this approach in producing diagnostic test sets whose numbers of tests are closer to those of compact fault detection test sets while preserving the accuracy of logic diagnosis.

ECFNet: A Siamese Network With Fewer FPs and Fewer FNs for Change Detection of Remote-Sensing Images

High-resolution remote-sensing image change detection plays an important role in areas such as land and resources investigation, natural disaster prediction, and military strategy research. Current change detection methods often focus on extracting more discriminative features, while ignoring the information loss and imbalance problems in the process of feature fusion, which results in weakness in small change objects and edge pixels of change objects. In this letter, a simple but efficient network architecture, extraction, comparison and fusion network (ECFNet), for change detection in remote-sensing images is proposed. By constraining the number of feature channels in the fusion process by the feature comparison module (FCM), ECFNet can better utilize the fine-grained information in the multiscale feature map for result prediction, which not only improves the detection performance of small objects, but also reduces the false detection around the edge pixels of change objects. Experiments on the deeply supervised image fusion network for change detection (DSIFN-CD) test set show that ECFNet achieves state-of-the-art results with a small amount of computation.

Learning From Synthetic InSAR With Vision Transformers: The Case of Volcanic Unrest Detection

The detection of early signs of volcanic unrest preceding an eruption, in the form of ground deformation in Interferometric Synthetic Aperture Radar (InSAR) data is critical for assessing volcanic hazard. In this work we treat this as a binary classification problem of InSAR images, and propose a novel deep learning methodology that exploits a rich source of synthetically generated interferograms to train quality classifiers that perform equally well in real interferograms. The imbalanced nature of the problem, with orders of magnitude fewer positive samples, coupled with the lack of a curated database with labeled InSAR data, sets a challenging task for conventional deep learning architectures. We propose a new framework for domain adaptation, in which we learn class prototypes from synthetic data with vision transformers. We report detection accuracy that amounts to the highest reported accuracy on a large test set for volcanic unrest detection. Moreover, we built upon this knowledge by learning a new, non-linear, projection between the learnt representations and prototype space, using pseudo labels produced by our model from an unlabeled real InSAR dataset. This leads to the new state of the art with 97.1% accuracy on our test set. We demonstrate the robustness of our approach by training a simple ResNet-18 Convolutional Neural Network on the unlabeled real InSAR dataset with pseudo-labels generated from our top transformer-prototype model. Our methodology provides a significant improvement in performance without the need of manually labeling any sample, opening the road for further exploitation of synthetic InSAR data in various remote sensing applications.

A fast and effective detection framework for whole-slide histopathology image analysis.

Pathologists generally pan, focus, zoom and scan tissue biopsies either under microscopes or on digital images for diagnosis. With the rapid development of whole-slide digital scanners for histopathology, computer-assisted digital pathology image analysis has attracted increasing clinical attention. Thus, the working style of pathologists is also beginning to change. Computer-assisted image analysis systems have been developed to help pathologists perform basic examinations. This paper presents a novel lightweight detection framework for automatic tumor detection in whole-slide histopathology images. We develop the Double Magnification Combination (DMC) classifier, which is a modified DenseNet-40 to make patch-level predictions with only 0.3 million parameters. To improve the detection performance of multiple instances, we propose an improved adaptive sampling method with superpixel segmentation and introduce a new heuristic factor, local sampling density, as the convergence condition of iterations. In postprocessing, we use a CNN model with 4 convolutional layers to regulate the patch-level predictions based on the predictions of adjacent sampling points and use linear interpolation to generate a tumor probability heatmap. The entire framework was trained and validated using the dataset from the Camelyon16 Grand Challenge and Hubei Cancer Hospital. In our experiments, the average AUC was 0.95 in the test set for pixel-level detection.

Fault Detection of Wind Turbine Pitch System Based on Multiclass Optimal Margin Distribution Machine

In response to the unbalanced sample categories and complex sample distribution of the operating data of the pitch system of the wind turbine generator system, this paper proposes a method for fault detection of the pitch system of the wind turbine generator system based on the multiclass optimal margin distribution machine. In this method, the power output of the wind turbine generator system is used as the main status parameter, and the operating data history of the wind turbine generator system in the wind power supervisory control and data acquisition (SCADA) system is subject to correlation analysis with the Pearson correlation coefficient, to eliminate the features that have low correlation with the power output status parameter. Secondary analysis is performed to the remaining features, thus reducing the number and complexity of samples. Datasets are divided into the training set for training of the multiclass optimal margin distribution machine fault detection model and test set for testing. Experimental verification was carried out with the operating data of one wind farm in China. Experimental results show that, compared with other support vector machines, the proposed method has higher fault detection accuracy and precision and lower false-negative rate and false-positive rate.

О ЕДИНИЧНЫХ ПРОВЕРЯЮЩИХ ТЕСТАХ ОТНОСИТЕЛЬНО ЗАМЕН ЭЛЕМЕНТОВ НА ИНВЕРТОРЫ

О ЕДИНИЧНЫХ ПРОВЕРЯЮЩИХ ТЕСТАХ ОТНОСИТЕЛЬНО ЗАМЕН ЭЛЕМЕНТОВ НА ИНВЕРТОРЫ

Integrated Deep Model for Face Detection and Landmark Localization From “In The Wild” Images

The tasks of face detection and landmark localisation are a key foundation for many facial analysis applications, while great advancements have been achieved in recent years there are still challenges to increase the precision of face detection. Within this paper, we present our novel method the Integrated Deep Model (IDM), fusing two state-of-the-art deep learning architectures, namely, Faster R-CNN and a stacked hourglass for improved face detection precision and accurate landmark localisation. Integration is achieved through the application of a novel optimisation function and is shown in experimental evaluation to increase accuracy of face detection specifically precision by reducing false positive detection’s by an average of 62%. Our proposed IDM method is evaluated on the Annotated Faces In-The-Wild, Annotated Facial Landmarks In The Wild and the Face Detection Dataset and Benchmark face detection test sets and shows a high level of recall and precision when compared with previously proposed methods. Landmark localisation is evaluated on the Annotated Faces In-The-Wild and 300-W test sets, this specifically focuses on localisation accuracy from detected face bounding boxes when compared with baseline evaluations using ground truth bounding boxes. Our findings highlight only a small 0.005% maximum increase in error which is more profound for the subset of facial landmarks which border the face.

COMEDI: Combinatorial Election of Diagnostic Vectors From Detection Test Sets for Logic Circuits

Although the modern automatic test pattern generation (ATPG) tools can efficiently produce near-optimal test sets with high fault-coverage for a circuit-under-test, a diagnostic test set (DTS), which is needed for fault localization, is much more challenging to construct. The DTS is used to analyze the responses of failing chips during manufacturing test for the purpose of identifying the root cause of observed errors. In this paper, a novel technique for selecting a powerful DTS for stuck-at faults from a pool of ATPG detection vectors is proposed. Unlike existing methods, this technique does not use any diagnostic test generation, circuit modification, or miter-based approach. It constructs a combinatorial cover of the pool to determine a test set with high diagnostic coverage (DC). Two variants of the covering algorithm are proposed based on this technique. The experimental results on several combinational and scan-based benchmark circuits demonstrate the effectiveness of our method in terms of the size of the DTS, DC, and CPU time.

DeepID-Net: Deformable Deep Convolutional Neural Networks for Object Detection.

In this paper, we propose deformable deep convolutional neural networks for generic object detection. This new deep learning object detection framework has innovations in multiple aspects. In the proposed new deep architecture, a new deformation constrained pooling (def-pooling) layer models the deformation of object parts with geometric constraint and penalty. A new pre-training strategy is proposed to learn feature representations more suitable for the object detection task and with good generalization capability. By changing the net structures, training strategies, adding and removing some key components in the detection pipeline, a set of models with large diversity are obtained, which significantly improves the effectiveness of model averaging. The proposed approach improves the mean averaged precision obtained by RCNN [16], which was the state-of-the-art, from 31% to 50.3% on the ILSVRC2014 detection test set. It also outperforms the winner of ILSVRC2014, GoogLeNet, by 6.1%. Detailed component-wise analysis is also provided through extensive experimental evaluation, which provides a global view for people to understand the deep learning object detection pipeline.

N -Detection Test Sets for Circuits with Multiple Independent Scan Chains

In a circuit with multiple independent scan chains, it is possible to operate groups of scan chains independently in functional or shift mode. This design-for-testability approach can be used to increase the quality of a test set. This article describes an N -detection test generation procedure for increasing the quality of a transition fault test set in such a circuit. The procedure uses the possibility of applying the same test, with the scan chains operating in different modes, to increase the numbers of detections without increasing the number of tests that need to be generated or stored on a tester. This results in reduced input storage requirements compared with a conventional N -detection test set and an increased number of applied tests. The increased quality of the test set is verified by its bridging fault coverage.

On the design of switching circuits admitting small detection test sets

It is established that, for an arbitrary nonconstant Boolean function f(x1,..., xn), there exist the following testable switching circuits that admit a single fault detection test set of cardinality O(n): (a) a three-pole switching circuit (with one input and two output poles) that realizes a system of functions (f, \(\bar f\)) and (b) a two-pole switching circuit that realizes a function f(x1,..., xn) ⊕ xn+1. It is also proved that almost all Boolean functions f(x1,..., xn) can be realized by two-pole switching circuits that admit small detection test sets (test sets of cardinality O(n)) under homogeneous faults of switches (closures or breakings).

Improving the Accuracy of Defect Diagnosis by Considering Fewer Tests

Experimental results indicate that the addition of diagnostic tests to a fault detection test set may sometimes result in a larger set of candidate faults. Experimental results also indicate that a defect diagnosis procedure does not require the complete observed response of a faulty chip in order to produce accurate diagnosis results. Motivated by these observations, this paper augments a defect diagnosis procedure with a process that removes from consideration tests whose effects on the results of diagnosis may be negative. The augmented procedure runs the underlying defect diagnosis procedure several times in order to decide which tests should be removed from consideration. Experimental results indicate that this results in smaller sets of candidate faults and improved accuracy of diagnosis.

Functional Broadside Tests for Multistep Defect Diagnosis

In a two-step defect diagnosis process, a fault detection test set is used for initial diagnosis to compute an initial set of candidate faults. When the initial set is large, diagnostic tests are generated based on the candidate faults in the set, and the set is refined based on the extended test set. This paper investigates the ability of functional broadside tests to serve as diagnostic tests for refining initial sets of candidate faults. The paper discusses the advantages of using functional broadside tests for this purpose. These advantages are related to the fact that the tests create functional operation conditions, and thus avoid nonfunctional effects that may make diagnosis less accurate. It also describes a multistep defect diagnosis process that uses functional broadside tests. Experimental results are presented to show the extent to which functional broadside tests can reduce initial sets of candidate faults.

Method of synthesis of easily testable circuits admitting single fault detection tests of constant length

It is constructively proved that any nonconstant Boolean function may be realized (over an arbitrary complete basis of gates) by a testable combinational circuit admitting (under single inverse or arbitrary constant faults at outputs of gates) a single fault detection test set whose cardinality does not exceed 4.

On the synthesis of circuits admitting complete fault detection test sets of constant length under arbitrary constant faults at the outputs of the gates

On the synthesis of circuits admitting complete fault detection test sets of constant length under arbitrary constant faults at the outputs of the gates

Derivation of test set for detecting multiple missing-gate faults in reversible circuits

Logic synthesis of reversible circuits has become an important problem because of its relevance to the emerging area of quantum computation. Special types of quantum gates such as k-CNOT may be deployed to implement a reversible circuit. Although the classical stuck-at fault model is widely used for modeling defects in conventional CMOS circuits, new approaches, namely single missing-gate fault (SMGF), repeated-gate fault (RGF), partial missing-gate fault (PMGF), and multiple missing-gate fault (MMGF), have been found to be more befitting for modeling defects in quantum k-CNOT gates. This article presents an algorithm to derive a test set (TS) for detection of multiple missing-gate faults in a reversible circuit implemented with k-CNOT gates. It is shown that TS is sufficient to detect all single missing-gate faults (SMGFs) and all detectable repeated gate faults (RGFs). Experimental results on test set for some benchmark circuits are reported, which compare favorably with earlier findings.

Non-Uniform Coverage by $n$-Detection Test Sets

The use of n-detection test sets increases the likelihood of defect detection. With a uniform value of for all the target faults, the expectation is that defects across the circuit will be covered uniformly. This paper demonstrates that this may not be the case by considering the four-way bridging faults detected by n-detection test sets for single stuck-at faults in benchmark circuits. Partitioning the bridging faults into subsets according to their dominated line, the results show that certain subsets have significantly lower bridging fault coverage than others. Thus, certain defect sites are significantly less covered than others. This paper also shows that it is possible to predict which subsets will have low bridging fault coverage based on the numbers of detections of single stuck-at faults under a conventional one-detection test set. This observation leads to a simplified n-detection test generation strategy. It also points to the possibility of using higher numbers of detections for certain single stuck-at faults, or targeting other fault models only at the sites that are expected to be less covered.

Multicycle Tests With Constant Primary Input Vectors for Increased Fault Coverage

Test generation procedures for n -detection test sets improve the quality of a test set by adding tests that increase the numbers of detections of target faults. A different approach to n-detection test generation increases the numbers of detections of target faults within the bounds of the number of tests of a single-detection test set. Multicycle tests provide the flexibility of improving the quality of a test set by increasing the number of clock cycles in each test, without increasing the number of tests. Improved test quality is thus achieved with limited increases in test application time and test data volume due to the larger numbers of clock cycles in each test. This paper describes a procedure that starts from a compact one-detection single-cycle test set for single stuck-at faults and produces a multicycle test set with the same number of tests, but increased numbers of clock cycles and improved test quality. The procedure uses only one-detection fault simulation of single stuck-at faults. A similar procedure is applied starting from a two-cycle test set and considering transition faults. The procedures produce tests with constant primary input vectors to accommodate tester limitations.

Multi-Pattern $n$-Detection Stuck-At Test Sets for Delay Defect Coverage

An n-detection test set detects each target fault n times. A higher value of n increases the likelihood that defects around the site of a target fault will be detected. However, an n -detection test set that consists of single-pattern tests (generated for single stuck-at faults) cannot guarantee that delay defects will be detected. For this it is necessary to use multi-pattern tests. The procedure described in this work generates a multi-pattern n-detection test set for single stuck-at faults. The procedure is applied starting from a single-pattern n-detection test set for single stuck-at faults. Experimental results demonstrate that the multi-pattern n-detection test set achieves a high transition fault coverage. It achieves a similar bridging fault coverage to the single-pattern test set. An added advantage is that the multi-pattern test set typically requires significantly fewer clock cycles for test application than the single-pattern test set.