Automated pavement distress detection via road images is still a challenging issue among pavement researchers and computer-vision community. In recent years, advancement in deep learning has enabled researchers to develop robust tools for analyzing pavement images at unprecedented accuracies. Nevertheless, deep learning models necessitate a big ground truth dataset, which is often not readily accessible for pavement field. In this study, we reviewed our previous study, which a labeled pavement dataset was presented as the first step towards a more robust, easy-to-deploy pavement condition assessment system. In total, 7237 google street-view images were extracted, manually annotated for classification (nine categories of distress classes). Afterward, YOLO (you look only once) deep learning framework was implemented to train the model using the labeled dataset. In the current study, a U-net based model is developed to quantify the severity of the distresses, and finally, a hybrid model is developed by integrating the YOLO and U-net model to classify the distresses and quantify their severity simultaneously. Various pavement condition indices are developed by implementing various machine learning algorithms using the YOLO deep learning framework for distress classification and U-net for segmentation and distress densification. The output of the distress classification and segmentation models are used to develop a comprehensive pavement condition tool which rates each pavement image according to the type and severity of distress extracted.
Although few-shot learning and one-class classification (OCC), i.e., learning a binary classifier with data from only one class, have been separately well studied, their intersection remains rather unexplored. Our work addresses the few-shot OCC problem and presents a method to modify the episodic data sampling strategy of the model-agnostic meta-learning (MAML) algorithm to learn a model initialization particularly suited for learning few-shot OCC tasks. This is done by explicitly optimizing for an initialization which only requires few gradient steps with one-class minibatches to yield a performance increase on class-balanced test data. We provide a theoretical analysis that explains why our approach works in the few-shot OCC scenario, while other meta-learning algorithms fail, including the unmodified MAML. Our experiments on eight datasets from the image and time-series domains show that our method leads to better results than classical OCC and few-shot classification approaches, and demonstrate the ability to learn unseen tasks from only few normal class samples. Moreover, we successfully train anomaly detectors for a real-world application on sensor readings recorded during industrial manufacturing of workpieces with a CNC milling machine, by using few normal examples. Finally, we empirically demonstrate that the proposed data sampling technique increases the performance of more recent meta-learning algorithms in few-shot OCC and yields state-of-the-art results in this problem setting.
Attention mechanism has been regarded as an advanced technique to capture long-range feature interactions and to boost the representation capability for convolutional neural networks. However, we found two ignored problems in current attentional activations-based models: the approximation problem and the insufficient capacity problem of the attention maps. To solve the two problems together, we initially propose an attention module for convolutional neural networks by developing an AW-convolution, where the shape of attention maps matches that of the weights rather than the activations. Our proposed attention module is a complementary method to previous attention-based schemes, such as those that apply the attention mechanism to explore the relationship between channel-wise and spatial features. Experiments on several datasets for image classification and object detection tasks show the effectiveness of our proposed attention module. In particular, our proposed attention module achieves 1.00% Top-1 accuracy improvement on ImageNet classification over a ResNet101 baseline and 0.63 COCO-style Average Precision improvement on the COCO object detection on top of a Faster R-CNN baseline with the backbone of ResNet101-FPN. When integrating with the previous attentional activations-based models, our proposed attention module can further increase their Top-1 accuracy on ImageNet classification by up to 0.57% and COCO-style Average Precision on the COCO object detection by up to 0.45. Code and pre-trained models will be publicly available.
For the MIDOG mitosis detection challenge, we created a cascade algorithm consisting of a Mask-RCNN detector, followed by a classification ensemble consisting of ResNet50 and DenseNet201 to refine detected mitotic candidates. The MIDOG training data consists of 200 frames originating from four scanners, three of which are annotated for mitotic instances with centroid annotations. Our main algorithmic choices are as follows: first, to enhance the generalizability of our detector and classification networks, we use a state-of-the-art residual Cycle-GAN to transform each scanner domain to every other scanner domain. During training, we then randomly load, for each image, one of the four domains. In this way, our networks can learn from the fourth non-annotated scanner domain even if we don't have annotations for it. Second, for training the detector network, rather than using centroid-based fixed-size bounding boxes, we create mitosis-specific bounding boxes. We do this by manually annotating a small selection of mitoses, training a Mask-RCNN on this small dataset, and applying it to the rest of the data to obtain full annotations. We trained the follow-up classification ensemble using only the challenge-provided positive and hard-negative examples. On the preliminary test set, the algorithm scores an F1 score of 0.7578, putting us as the second-place team on the leaderboard.
Modern radio telescopes will daily generate data sets on the scale of exabytes for systems like the Square Kilometre Array (SKA). Massive data sets are a source of unknown and rare astrophysical phenomena that lead to discoveries. Nonetheless, this is only plausible with the exploitation of intensive machine intelligence to complement human-aided and traditional statistical techniques. Recently, there has been a surge in scientific publications focusing on the use of artificial intelligence in radio astronomy, addressing challenges such as source extraction, morphological classification, and anomaly detection. This study presents a succinct, but comprehensive review of the application of machine intelligence techniques on radio images with emphasis on the morphological classification of radio galaxies. It aims to present a detailed synthesis of the relevant papers summarizing the literature based on data complexity, data pre-processing, and methodological novelty in radio astronomy. The rapid advancement and application of computer intelligence in radio astronomy has resulted in a revolution and a new paradigm shift in the automation of daunting data processes. However, the optimal exploitation of artificial intelligence in radio astronomy, calls for continued collaborative efforts in the creation of annotated data sets. Additionally, in order to quickly locate radio galaxies with similar or dissimilar physical characteristics, it is necessary to index the identified radio sources. Nonetheless, this issue has not been adequately addressed in the literature, making it an open area for further study.
Large-scale pre-trained Vision Language Models (VLMs) have proven effective for zero-shot classification. Despite the success, most traditional VLMs-based methods are restricted by the assumption of partial source supervision or ideal vocabularies, which rarely satisfy the open-world scenario. In this paper, we aim at a more challenging setting, Realistic Zero-Shot Classification, which assumes no annotation but instead a broad vocabulary. To address this challenge, we propose the Self Structural Semantic Alignment (S^3A) framework, which extracts the structural semantic information from unlabeled data while simultaneously self-learning. Our S^3A framework adopts a unique Cluster-Vote-Prompt-Realign (CVPR) algorithm, which iteratively groups unlabeled data to derive structural semantics for pseudo-supervision. Our CVPR process includes iterative clustering on images, voting within each cluster to identify initial class candidates from the vocabulary, generating discriminative prompts with large language models to discern confusing candidates, and realigning images and the vocabulary as structural semantic alignment. Finally, we propose to self-learn the CLIP image encoder with both individual and structural semantic alignment through a teacher-student learning strategy. Our comprehensive experiments across various generic and fine-grained benchmarks demonstrate that the S^3A method offers substantial improvements over existing VLMs-based approaches, achieving a more than 15% accuracy improvement over CLIP on average. Our codes, models, and prompts are publicly released at https://github.com/sheng-eatamath/S3A.
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.
Convolutional Neural Networks (CNN) have revolutionized perception for color images, and their application to sonar images has also obtained good results. But in general CNNs are difficult to train without a large dataset, need manual tuning of a considerable number of hyperparameters, and require many careful decisions by a designer. In this work, we evaluate three common decisions that need to be made by a CNN designer, namely the performance of transfer learning, the effect of object/image size and the relation between training set size. We evaluate three CNN models, namely one based on LeNet, and two based on the Fire module from SqueezeNet. Our findings are: Transfer learning with an SVM works very well, even when the train and transfer sets have no classes in common, and high classification performance can be obtained even when the target dataset is small. The ADAM optimizer combined with Batch Normalization can make a high accuracy CNN classifier, even with small image sizes (16 pixels). At least 50 samples per class are required to obtain $90\%$ test accuracy, and using Dropout with a small dataset helps improve performance, but Batch Normalization is better when a large dataset is available.
Developing a suitable Deep Neural Network (DNN) often requires significant iteration, where different model versions are evaluated and compared. While metrics such as accuracy are a powerful means to succinctly describe a model's performance across a dataset or to directly compare model versions, practitioners often wish to gain a deeper understanding of the factors that influence a model's predictions. Interpretability techniques such as gradient-based methods and local approximations can be used to examine small sets of inputs in fine detail, but it can be hard to determine if results from small sets generalize across a dataset. We introduce IMACS, a method that combines gradient-based model attributions with aggregation and visualization techniques to summarize differences in attributions between two DNN image models. More specifically, IMACS extracts salient input features from an evaluation dataset, clusters them based on similarity, then visualizes differences in model attributions for similar input features. In this work, we introduce a framework for aggregating, summarizing, and comparing the attribution information for two models across a dataset; present visualizations that highlight differences between 2 image classification models; and show how our technique can uncover behavioral differences caused by domain shift between two models trained on satellite images.
Recently proposed large-scale text-to-image generative models such as DALL$\cdot$E 2, Midjourney, and StableDiffusion can generate high-quality and realistic images from users' prompts. Not limited to the research community, ordinary Internet users enjoy these generative models, and consequently, a tremendous amount of generated images have been shared on the Internet. Meanwhile, today's success of deep learning in the computer vision field owes a lot to images collected from the Internet. These trends lead us to a research question: "\textbf{will such generated images impact the quality of future datasets and the performance of computer vision models positively or negatively?}" This paper empirically answers this question by simulating contamination. Namely, we generate ImageNet-scale and COCO-scale datasets using a state-of-the-art generative model and evaluate models trained with "contaminated" datasets on various tasks, including image classification and image generation. Throughout experiments, we conclude that generated images negatively affect downstream performance, while the significance depends on tasks and the amount of generated images. The generated datasets and the codes for experiments will be publicly released for future research. Generated datasets and source codes are available from \url{https://github.com/moskomule/dataset-contamination}.
본 논문에서는 확산 모델 내 클래스 빈도와 다중 스케일 노이즈 스케줄 간의 상관관계를 최초로 조사한다. 점수 기반(score-based) 생성 모델에서 저밀도 영역은 점수가 부정확하게 추정되는 경우가 많아 생성 품질을 저하시킨다. 다중 스케일 노이즈 스케줄이 확산 과정에서 이 문제를 완화할 수 있지만, 저빈도 클래스는 여전히 큰 저밀도 영역에 직면하여 고빈도 클래스보다 점수 추정이 더 부정확해지는 문제에 직면한다. 또한 고빈도 클래스가 점수 공간을 주도하는 경향이 있어, 대부분의 데이터 포인트가 이러한 클래스의 샘플을 생성하는 방향으로 수렴한다. 그 결과, 저빈도 클래스 내에서 생성된 샘플은 품질이 최적이 아니고 다양성이 제한적이다. 이 문제를 해결하기 위해, 저빈도 클래스에 더 큰 스케일의 노이즈를 부여해야 한다는 통찰을 활용한 클래스 빈도 유도(Class-frequency Guided, CFRG) 노이즈 스케줄을 제안한다. 우리 방법의 효과를 보여주기 위해, CIFAR-100-LT 및 ImageNet-LT 등 불균형 데이터셋을 사용하여 이미지 생성, 이미지 분류, 텍스트-이미지 생성 등 다양한 작업에서 실험을 수행한다. CFRG 노이즈 스케줄을 적용하여 기준선 대비 상당한 향상을 달성하고, 노이즈 스케줄 설계에서 빈도 통계의 핵심적인 역할을 보여준다.
바이트 공간 동작 인식(Byte-space Action Recognition, BAR)은 픽셀 디코딩 없이 압축된 이미지 비트스트림에서 직접 인간의 동작을 인식하는 것을 목표로 한다. 전적으로 바이트 공간에서 작동함으로써 BAR은 본질적으로 파일 무결성 및 픽셀 수준의 재구성에 독립적이므로, 프라이버시에 민감한 시나리오에 자연스럽게 적용할 수 있고 비트스트림 손상에 견고하다. 본 논문에서는 손상된 이미지 비트스트림에서도 정확한 동작 인식을 달성하는 BAR 기반 모델인 ByteAction을 제안한다. ByteAction은 듀얼 뷰(dual-view) 바이트 수준 인식 프레임워크를 따른다. 이 모델은 비트스트림 패턴 증강(Bitstream Pattern Augmentation, BPA)을 통해 증강되고 공유된 ByteFormer 백본으로 인코딩되는 약하게 손상된 비트스트림 뷰와 강하게 손상된 비트스트림 뷰를 구성한다. 해당 모델은 분류 및 손상 일관성 목적 함수 모두를 통해 최적화된다. 구체적으로, 1차원 바이트 시퀀스를 2차원 바이트 행렬로 재구성하고 영역 수준 삭제를 적용하여 모델이 견고한 영역 간 바이트 의존성을 학습하도록 유도하는 비트스트림 패턴 증강(BPA)을 제안한다. 나아가 양방향 KL 발산(bidirectional KL divergence)을 통해 모델이 다양한 손상 심각도 전반에서 안정적인 예측을 유지하도록 제한하는 손상 일관성 학습(Corruption Consistency Training) 전략을 제안한다. Stanford40, PPMI, PASCAL VOC 2012 Action의 이미지 비트스트림에 대한 실험은 ByteAction이 손상되지 않은 비트스트림 성능을 경쟁력 있게 유지하면서 모든 시나리오에서 최고 수준의 손상 견고성을 달성함을 입증한다.
While most previous automation-assisted reading methods can improve efficiency, their performance often relies on the success of accurate cell segmentation and hand-craft feature extraction. This paper presents an efficient and totally segmentation-free method for automated cervical cell screening that utilizes modern object detector to directly detect cervical cells or clumps, without the design of specific hand-crafted feature. Specifically, we use the state-of-the-art CNN-based object detection methods, YOLOv3, as our baseline model. In order to improve the classification performance of hard examples which are four highly similar categories, we cascade an additional task-specific classifier. We also investigate the presence of unreliable annotations and cope with them by smoothing the distribution of noisy labels. We comprehensively evaluate our methods on test set which is consisted of 1,014 annotated cervical cell images with size of 4000*3000 and complex cellular situation corresponding to 10 categories. Our model achieves 97.5% sensitivity (Sens) and 67.8% specificity (Spec) on cervical cell image-level screening. Moreover, we obtain a mean Average Precision (mAP) of 63.4% on cervical cell-level diagnosis, and improve the Average Precision (AP) of hard examples which are valuable but difficult to distinguish. Our automation-assisted cervical cell reading method not only achieves cervical cell image-level classification but also provides more detailed location and category information of abnormal cells. The results indicate feasible performance of our method, together with the efficiency and robustness, providing a new idea for future development of computer-assisted reading system in clinical cervical screening.
Volume Rendering is an important technique for visualizing three-dimensional scalar data grids and is commonly employed for scientific and medical image data. Direct Volume Rendering (DVR) is a well established and efficient rendering algorithm for volumetric data. Neural rendering uses deep neural networks to solve inverse rendering tasks and applies techniques similar to DVR. However, it has not been demonstrated successfully for the rendering of scientific volume data. In this work, we introduce Deep Direct Volume Rendering (DeepDVR), a generalization of DVR that allows for the integration of deep neural networks into the DVR algorithm. We conceptualize the rendering in a latent color space, thus enabling the use of deep architectures to learn implicit mappings for feature extraction and classification, replacing explicit feature design and hand-crafted transfer functions. Our generalization serves to derive novel volume rendering architectures that can be trained end-to-end directly from examples in image space, obviating the need to manually define and fine-tune multidimensional transfer functions while providing superior classification strength. We further introduce a novel stepsize annealing scheme to accelerate the training of DeepDVR models and validate its effectiveness in a set of experiments. We validate our architectures on two example use cases: (1) learning an optimized rendering from manually adjusted reference images for a single volume and (2) learning advanced visualization concepts like shading and semantic colorization that generalize to unseen volume data. We find that deep volume rendering architectures with explicit modeling of the DVR pipeline effectively enable end-to-end learning of scientific volume rendering tasks from target images.
There exist a wide range of single number metrics for assessing performance of classification algorithms, including AUC and the F1-score (Wikipedia lists 17 such metrics, with 27 different names). In this article, I propose a new metric to answer the following question: when an algorithm is tuned so that it can no longer distinguish labelled cats from real cats, how often does a randomly chosen image that has been labelled as containing a cat actually contain a cat? The steps to construct this metric are as follows. First, we set a threshold score such that when the algorithm is shown two randomly-chosen images -- one that has a score greater than the threshold (i.e. a picture labelled as containing a cat) and another from those pictures that really does contain a cat -- the probability that the image with the highest score is the one chosen from the set of real cat images is 50\%. At this decision threshold, the set of positively labelled images are indistinguishable from the set of images which are positive. Then, as a second step, we measure performance by asking how often a randomly chosen picture from those labelled as containing a cat actually contains a cat. This metric can be thought of as {\it precision at the indistinguishability threshold}. While this new metric doesn't address the tradeoff between precision and recall inherent to all such metrics, I do show why this method avoids pitfalls that can occur when using, for example AUC, and it is better motivated than, for example, the F1-score.
We propose a novel algorithm for weakly supervised semantic segmentation based on image-level class labels only. In weakly supervised setting, it is commonly observed that trained model overly focuses on discriminative parts rather than the entire object area. Our goal is to overcome this limitation with no additional human intervention by retrieving videos relevant to target class labels from web repository, and generating segmentation labels from the retrieved videos to simulate strong supervision for semantic segmentation. During this process, we take advantage of image classification with discriminative localization technique to reject false alarms in retrieved videos and identify relevant spatio-temporal volumes within retrieved videos. Although the entire procedure does not require any additional supervision, the segmentation annotations obtained from videos are sufficiently strong to learn a model for semantic segmentation. The proposed algorithm substantially outperforms existing methods based on the same level of supervision and is even as competitive as the approaches relying on extra annotations.
Neural networks have proven effective at solving difficult problems but designing their architectures can be challenging, even for image classification problems alone. Our goal is to minimize human participation, so we employ evolutionary algorithms to discover such networks automatically. Despite significant computational requirements, we show that it is now possible to evolve models with accuracies within the range of those published in the last year. Specifically, we employ simple evolutionary techniques at unprecedented scales to discover models for the CIFAR-10 and CIFAR-100 datasets, starting from trivial initial conditions and reaching accuracies of 94.6% (95.6% for ensemble) and 77.0%, respectively. To do this, we use novel and intuitive mutation operators that navigate large search spaces; we stress that no human participation is required once evolution starts and that the output is a fully-trained model. Throughout this work, we place special emphasis on the repeatability of results, the variability in the outcomes and the computational requirements.
Convolutional Neural Networks (CNNs) have been successfully applied to many computer vision tasks, such as image classification. By performing linear combinations and element-wise nonlinear operations, these networks can be thought of as extracting solely first-order information from an input image. In the past, however, second-order statistics computed from handcrafted features, e.g., covariances, have proven highly effective in diverse recognition tasks. In this paper, we introduce a novel class of CNNs that exploit second-order statistics. To this end, we design a series of new layers that (i) extract a covariance matrix from convolutional activations, (ii) compute a parametric, second-order transformation of a matrix, and (iii) perform a parametric vectorization of a matrix. These operations can be assembled to form a Covariance Descriptor Unit (CDU), which replaces the fully-connected layers of standard CNNs. Our experiments demonstrate the benefits of our new architecture, which outperform the first-order CNNs, while relying on up to 90% fewer parameters.
Deep Neural Networks have achieved extraordinary results on image classification tasks, but have been shown to be vulnerable to attacks with carefully crafted perturbations of the input data. Although most attacks usually change values of many image's pixels, it has been shown that deep networks are also vulnerable to sparse alterations of the input. However, no computationally efficient method has been proposed to compute sparse perturbations. In this paper, we exploit the low mean curvature of the decision boundary, and propose SparseFool, a geometry inspired sparse attack that controls the sparsity of the perturbations. Extensive evaluations show that our approach computes sparse perturbations very fast, and scales efficiently to high dimensional data. We further analyze the transferability and the visual effects of the perturbations, and show the existence of shared semantic information across the images and the networks. Finally, we show that adversarial training can only slightly improve the robustness against sparse additive perturbations computed with SparseFool.
Transfer learning from natural image datasets, particularly ImageNet, using standard large models and corresponding pretrained weights has become a de-facto method for deep learning applications to medical imaging. However, there are fundamental differences in data sizes, features and task specifications between natural image classification and the target medical tasks, and there is little understanding of the effects of transfer. In this paper, we explore properties of transfer learning for medical imaging. A performance evaluation on two large scale medical imaging tasks shows that surprisingly, transfer offers little benefit to performance, and simple, lightweight models can perform comparably to ImageNet architectures. Investigating the learned representations and features, we find that some of the differences from transfer learning are due to the over-parametrization of standard models rather than sophisticated feature reuse. We isolate where useful feature reuse occurs, and outline the implications for more efficient model exploration. We also explore feature independent benefits of transfer arising from weight scalings.