Hyperorthogonal well-folded Hilbert curves

Supervised object detection models are trained to recognize certain objects. These models are classified into two types: single-stage detectors and two-stage detectors. The single-stage detectors just need one pass through the model to anticipate all the bounding boxes, whereas the two-stage detectors require to first estimate the image portions where the object could be located. Due to their speed and simplicity, single-stage anchor-based models are used in many industrial settings. Training such models require bounding boxes that describe the spatial location of an object, which are usually drawn by an expert. However, the question remains, how much area should be considered when drawing the bounding boxes? In this paper, we demonstrate the effects that the size and placement of a rectangular bounding box can have on the performance of the anchor-based models. For this, we first perform experiments on a synthetically generated binary dataset and then on a real-world object detection dataset. Our results show that fixing the size of the bounding boxes can help in improving the performance of the model in the case of single class object detection (approximately 50% improvement in mAP@[.5:.95] for real world dataset). Furthermore, we also demonstrate how freely available tools can be combined for obtaining the best possible semi automated object labeling pipeline.

Efficiently querying large trajectory datasets is a challenge of growing importance. Abstracting trajectory segments with minimum bounding boxes and indexing them in R-Trees results in a high false positive rate due to high dead space. Space filling curves (SFCs), which have excellent locality preserving and dimensionality reduction properties, have been shown to be effective for indexing points in space. However, they can yield a high false positive count and slow query times if used to index trajectory segments. Our work shows how to use SFCs to index trajectory polylines. In our experiments, the proposed method runs 2--15 times faster than other state-of-the-art approaches.

The complex underwater environment and lighting conditions make underwater images suffer from texture distortion and color variations. In this paper, we propose an improved YOLOv4 detection method to detect four underwater organisms: holothurian, echinus, scallop, starfish and waterweeds. Firstly, we modified the network structure, added a deep separable convolution to the backbone network, and added a 152×152 feature map, which is conducive to the detection of small targets. Secondly, k-means clustering algorithm is used to cluster the bounding box in the data set, and the size of the bounding box is improved according to the clustering results. Thirdly, we propose a new module (EASPP, Spatial Pyramid Pooling), which increases slightly the model complexity, but the improvement effect is significant. Finally, when training the model, we use multi-scale training to better train targets with different scales. The experimental results show that on our test set, the improved method in the underwater object detection method is 4.8% higher than the original YOLOv4 model in accuracy (AP), the F1-score is 5.1% higher than that of the original method, and for mAP@0.5 it reaches 81.5%, which is 5.6% higher than that of the original method, which can be concluded that our method is effective.

We introduce a new large-scale data set of video URLs with densely-sampled object bounding box annotations called YouTube-BoundingBoxes (YT-BB). The data set consists of approximately 380,000 video segments about 19s long, automatically selected to feature objects in natural settings without editing or post-processing, with a recording quality often akin to that of a hand-held cell phone camera. The objects represent a subset of the COCO [32] label set. All video segments were human-annotated with high-precision classification labels and bounding boxes at 1 frame per second. The use of a cascade of increasingly precise human annotations ensures a label accuracy above 95% for every class and tight bounding boxes. Finally, we train and evaluate well-known deep network architectures and report baseline figures for per-frame classification and localization. We also demonstrate how the temporal contiguity of video can potentially be used to improve such inferences. The data set can be found at https://research.google.com/youtube-bb. We hope the availability of such large curated corpus will spur new advances in video object detection and tracking.

The non-maximum suppression (NMS) algorithm, which merges neighboring bounding boxes as the detection result, is widely utilized in object detection methods. However, the traditional NMS algorithm is not suitable for text detection in natural scenes, especially for long texts and dense texts. In this paper, we observe that the coordinates of candidate bounding boxes are presented in skew distribution. On this observation, an improved NMS algorithm called SD-NMS (Skew Distribution NMS) is designed. First, the mode and the median of the coordinate set are counted to filter out the redundant bounding boxes. Then the left bounding boxes are merged for the location of the text regions. The SD-NMS method improves the detection ability of the model without extra model training and can be easily embedded in text detection methods. The experimental results show that our method obtains F-measure more than other NMS methods in public data sets ICDAR2015 and MSRATD500.

In this paper, we introduce the outlier detection problem in a set of uncertain points. We study two variants of the problems based upon the definition of the outlier. For a given positive integer \(k(<n)\) and a set \(\mathfrak {R}\) of n regions as the imprecise points, the first type of the outlier detection problem that we study is to locate \(n-k\) points on distinct regions, such that the size of the smallest axis-aligned bounding box (AABB), the diameter or the smallest enclosing circle (SEC) of the resulting points gets minimized. The uncertainty regions we study are squares or disks, and the excluded k regions are considered as outliers.We also study the covering versions in which the objectives of the SEC and the AABB problems are to find the smallest circle or axis-aligned bounding box, respectively, that covers the area of at least \(n-k\) regions.In the second-type of outliers, the outliers are those k regions that mostly reduce the uncertainty-induced gap between the lower bound and the upper bound on the size of the output. We give polynomial time algorithms for several variants of the mentioned problems, ranging in running time from \(O(n \log n)\) to \(O(n^{5.5}\log n)\).

Automatic localization of pneumonia on chest X-rays (CXRs) is highly desirable both as an interpretive aid to the radiologist and for timely diagnosis of the disease. However, pneumonia's amorphous appearance on CXRs and complexity of normal anatomy in the chest present key challenges that hinder accurate localization. Existing studies in this area are either not optimized to preserve spatial information of abnormality or depend on expensive expert-annotated bounding boxes. We present a novel generative adversarial network (GAN)-based machine learning approach for this problem, which is weakly supervised (does not require any location annotations), was trained to retain spatial information, and can produce pixel-wise abnormality maps highlighting regions of abnormality (as opposed to bounding boxes around abnormality). Our method is based on the Wasserstein GAN framework and, to the best of our knowledge, the first application of GANs to this problem. Specifically, from an abnormal CXR as input, we generated the corresponding pseudo normal CXR image as output. The pseudo normal CXR is the "hypothetical" normal, if the same abnormal CXR were not to have any abnormalities. We surmise that the difference between the pseudo normal and the abnormal CXR highlights the pixels suspected to have pneumonia and hence is our output abnormality map. We trained our algorithm on an "unpaired" data set of abnormal and normal CXRs and did not require any location annotations such as bounding boxes/segmentations of abnormal regions. Furthermore, we incorporated additional prior knowledge/constraints into the model and showed that they help improve localization performance. We validated the model on a data set consisting of 14184 CXRs from the Radiological Society of North America pneumonia detectionchallenge. We evaluated our methods by comparing the generated abnormality maps with radiologist annotated bounding boxes using receiver operating characteristic (ROC) analysis, image similarity metrics such as normalized cross-correlation/mutual information, and abnormality detection rate.We also present visual examples of the abnormality maps, covering various scenarios of abnormality occurrence. Results demonstrate the ability to highlight regions of abnormality with the best method achieving an ROC area under the curve (AUC) of 0.77 and a detection rate of 85%.The GAN tended to perform better as prior knowledge/constraints were incorporated into themodel. We presented a novel GAN based approach for localizing pneumonia on CXRs that (1) does not require expensive hand annotated location ground truth; and (2) was trained to produce abnormality maps at the pixel level as opposed to bounding boxes. We demonstrated the efficacy of our methods via quantitative and qualitative results.

A performance evaluation protocol for the layout analysis is discussed in this paper. In the University of Washington English Document Image Database-III, there are 1600 English document images that come with manually edited ground truth of entity bounding boxes. These bounding boxes enclose text and non-text zones, text-lines, and words. We describe a performance metric for the comparison of the detected entities and the ground truth in terms of their bounding boxes. The Document Attribute Format Specification is used as the standard data representation. The protocol is intended to serve as a model for using the UW-III database to evaluate the document analysis algorithms. A set of layout analysis algorithms which detect different entities have been tested based on the data set and the performance metric. The evaluation results are presented in this paper.

Three dimensional point cloud data sets are easy to acquire and manipulate, but are often too large to process directly for embedded real-time applications. The spatial information in a point cloud can be represented in a variety of reduced forms, such as voxel grids, Gaussian mixture models, or spatial semantic structures. In this article, we show how a segmented point cloud can be represented as a spatial relationship graph using bounding boxes and triangular fuzzy numbers. This model is a lightweight encoding of the relative distance and direction between objects, and can be used to describe and query for particular spatial configurations using linguistic terms in a multicriteria framework. We show how this approach can be applied on a hand-segmented subset of the NPM3D data set with several illustrative examples. The work herein has useful applications in many applied domains, such as human-robot interaction with unmanned aerial systems.

In recent years, in the field of remote sensing image target detection, oriented target detection has received extensive attention. Traditional detectors use horizontal bounding box (HBB) to locate targets, which may cause inaccurate detection when detecting targets with arbitrary orientation angles, dense distribution and large aspect ratios. The Oriented Bounding Box (OBB) adds different rotation angles to the horizontal bounding box, which can better deal with the above problems. However, with the introduction of oriented bounding boxes in the detector, new problems have emerged, such as the increase in the number of anchor points with different scales, angles, and aspect ratios, and the imbalance of IoU matching of positive and negative anchor boxes, which leads to classification confidence and Inconsistency between positioning accuracy. In order to overcome these shortcomings while using the oriented bounding box, this paper proposes a novel oriented target detector and redesigns a matching strategy for evaluating the position of the anchor frame. By selecting a higher quality anchor frame to achieve more accurate target detection, thereby reducing the instability of the angle in the matching process. Experiments conducted on public remote sensing data sets such as HRSC2016 and UCAS-AOD show that the proposed method achieves advanced detection accuracy.

This paper mainly studies on the potential safety hazards in the obstacle recognition and processing system (ORPS) of the self-driving cars, which is constructed by deep learning architecture. We perform an attack that embeds a backdoor in the Mask R-CNN in ORPS by poisoning the dataset. Under normal circumstances, the backdoored model can accurately identify obstacles (vehicles). However, under certain circumstances, triggering the backdoor in the backdoored model may lead to change the size (bounding box and mask) and confidence of the detected obstacles, which may cause serious accidents. The experiment result shows that it is possible to embed a backdoor in ORPS. We can see that the backdoored network can obviously change the size of bounding box and corresponding mask of those poisoned instances. But on the other hand, embedding a backdoor in the deep learning based model will only slightly affect the accuracy of detecting objects without backdoor triggers, which is imperceptible for users. Eventually, we hope that our simple work can arouse people’s attention to the self-driving technology and even other deep learning based models. It brings motivation about how to judge or detect the existence of the backdoors in these systems.

BackgroundRecently, the Turing test has been used to investigate whether machines have intelligence similar to humans. Our study aimed to assess the ability of an artificial intelligence (AI) system for spine tumor detection using the Turing test.MethodsOur retrospective study data included 12179 images from 321 patients for developing AI detection systems and 6635 images from 187 patients for the Turing test. We utilized a deep learning-based tumor detection system with Faster R-CNN architecture, which generates region proposals by Region Proposal Network in the first stage and corrects the position and the size of the bounding box of the lesion area in the second stage. Each choice question featured four bounding boxes enclosing an identical tumor. Three were detected by the proposed deep learning model, whereas the other was annotated by a doctor; the results were shown to six doctors as respondents. If the respondent did not correctly identify the image annotated by a human, his answer was considered a misclassification. If all misclassification rates were >30%, the respondents were considered unable to distinguish the AI-detected tumor from the human-annotated one, which indicated that the AI system passed the Turing test.ResultsThe average misclassification rates in the Turing test were 51.2% (95% CI: 45.7%–57.5%) in the axial view (maximum of 62%, minimum of 44%) and 44.5% (95% CI: 38.2%–51.8%) in the sagittal view (maximum of 59%, minimum of 36%). The misclassification rates of all six respondents were >30%; therefore, our AI system passed the Turing test.ConclusionOur proposed intelligent spine tumor detection system has a similar detection ability to annotation doctors and may be an efficient tool to assist radiologists or orthopedists in primary spine tumor detection.

AD-YOLOv5: An object detection approach for key parts of sika deer based on deep learning

This report will cover the testing of the Ballistic Research Laboratory-Computer-Aided Design's (BRL-CAD's) ray-tracer on the primitive Bag of Triangles (BoTs). Two programs were created to accommodate each component of this project. The first program created all possible triangles (BoTs) given a bounding box and step size that is specified by the user. The second program fired multiple rays toward various locations on each triangle in the database. The results of these tests were classified as a success or failure; the percentages of success were graphed against the different tolerance levels to determine the accuracy of the ray-tracer on BoTs. These data could be useful in understanding how BRL-CAD's ray-tracer performs in comparison to the Air Force's ray-tracer, FASTGEN, on BoTs.

People head detection in crowded scenes is challenging due to the large variability in clothing and appearance, small scales of people, and strong partial occlusions. Traditional bottom-up proposal methods and existing region proposal network approaches suffer from either poor recall or low precision. In this paper, we propose to improve both the recall and precision of head detection of region proposal models by integrating the local head information. In specific, we first use a region proposal network to predict the bounding boxes and corresponding scores of multiple instances in the region. A local head classifier network is then trained to score the bounding box generated from the region proposal model. After that, we propose an adaptive fusion method by optimally combining both the region and local scores to obtain the final score of each candidate bounding box. Furthermore, our fusion models can automatically learn the optimal hyper-parameters from data. Our algorithm achieves superior people head detection performance on the crowded scenes data set, which significantly outperforms several recent state-of-the-art baselines in the literature.