We introduce ensembles of stochastic neural networks to approximate the Bayesian posterior, combining stochastic methods such as dropout with deep ensembles. The stochastic ensembles are formulated as families of distributions and trained to approximate the Bayesian posterior with variational inference. We implement stochastic ensembles based on Monte Carlo dropout, DropConnect and a novel non-parametric version of dropout and evaluate them on a toy problem and CIFAR image classification. For both tasks, we test the quality of the posteriors directly against Hamiltonian Monte Carlo simulations. Our results show that stochastic ensembles provide more accurate posterior estimates than other popular baselines for Bayesian inference.
This paper presents a supervised mixing augmentation method termed SuperMix, which exploits the salient regions within input images to construct mixed training samples. SuperMix is designed to obtain mixed images rich in visual features and complying with realistic image priors. To enhance the efficiency of the algorithm, we develop a variant of the Newton iterative method, $65\times$ faster than gradient descent on this problem. We validate the effectiveness of SuperMix through extensive evaluations and ablation studies on two tasks of object classification and knowledge distillation. On the classification task, SuperMix provides comparable performance to the advanced augmentation methods, such as AutoAugment and RandAugment. In particular, combining SuperMix with RandAugment achieves 78.2\% top-1 accuracy on ImageNet with ResNet50. On the distillation task, solely classifying images mixed using the teacher's knowledge achieves comparable performance to the state-of-the-art distillation methods. Furthermore, on average, incorporating mixed images into the distillation objective improves the performance by 3.4\% and 3.1\% on CIFAR-100 and ImageNet, respectively. {\it The code is available at https://github.com/alldbi/SuperMix}.
Neural Tensor Networks (NTNs), which are structured to encode the degree of relationship among pairs of entities, are used in Logic Tensor Networks (LTNs) to facilitate Statistical Relational Learning (SRL) in first-order logic. In this paper, we propose Randomly Weighted Tensor Networks (RWTNs), which incorporate randomly drawn, untrained tensors into an NTN encoder network with a trained decoder network. We show that RWTNs meet or surpass the performance of traditionally trained LTNs for Semantic Image Interpretation (SII) tasks that have been used as a representative example of how LTNs utilize reasoning over first-order logic to exceed the performance of solely data-driven methods. We demonstrate that RWTNs outperform LTNs for the detection of the relevant part-of relations between objects, and we show that RWTNs can achieve similar performance as LTNs for object classification while using fewer parameters for learning. Furthermore, we demonstrate that because the randomized weights do not depend on the data, several decoder networks can share a single NTN, giving RWTNs a unique economy of spatial scale for simultaneous classification tasks.
A long-standing issue with deep learning is the need for large and consistently labeled datasets. Although the current research in semi-supervised learning can decrease the required amount of annotated data by a factor of 10 or even more, this line of research still uses distinct classes like cats and dogs. However, in the real-world we often encounter problems where different experts have different opinions, thus producing fuzzy labels. We propose a novel framework for handling semi-supervised classifications of such fuzzy labels. Our framework is based on the idea of overclustering to detect substructures in these fuzzy labels. We propose a novel loss to improve the overclustering capability of our framework and show on the common image classification dataset STL-10 that it is faster and has better overclustering performance than previous work. On a real-world plankton dataset, we illustrate the benefit of overclustering for fuzzy labels and show that we beat previous state-of-the-art semisupervised methods. Moreover, we acquire 5 to 10% more consistent predictions of substructures.
In recent years, semi-supervised learning has been widely explored and shows excellent data efficiency for 2D data. There is an emerging need to improve data efficiency for 3D tasks due to the scarcity of labeled 3D data. This paper explores how the coherence of different modelities of 3D data (e.g. point cloud, image, and mesh) can be used to improve data efficiency for both 3D classification and retrieval tasks. We propose a novel multimodal semi-supervised learning framework by introducing instance-level consistency constraint and a novel multimodal contrastive prototype (M2CP) loss. The instance-level consistency enforces the network to generate consistent representations for multimodal data of the same object regardless of its modality. The M2CP maintains a multimodal prototype for each class and learns features with small intra-class variations by minimizing the feature distance of each object to its prototype while maximizing the distance to the others. Our proposed framework significantly outperforms all the state-of-the-art counterparts for both classification and retrieval tasks by a large margin on the modelNet10 and ModelNet40 datasets.
Federated learning (FL) is an appealing paradigm that allows a group of machines (a.k.a. clients) to learn collectively while keeping their data local. However, due to the heterogeneity between the clients' data distributions, the model obtained through the use of FL algorithms may perform poorly on some client's data. Personalization addresses this issue by enabling each client to have a different model tailored to their own data while simultaneously benefiting from the other clients' data. We consider an FL setting where some clients can be adversarial, and we derive conditions under which full collaboration fails. Specifically, we analyze the generalization performance of an interpolated personalized FL framework in the presence of adversarial clients, and we precisely characterize situations when full collaboration performs strictly worse than fine-tuned personalization. Our analysis determines how much we should scale down the level of collaboration, according to data heterogeneity and the tolerable fraction of adversarial clients. We support our findings with empirical results on mean estimation and binary classification problems, considering synthetic and benchmark image classification datasets.
초분광 이미지 분류는 여전히 단일 장면 내 무작위 픽셀 분할에 크게 의존한다. 무작위로 분할된 Salinas 데이터셋은 서로 다른 아키텍처를 비교하는 데 가장 널리 사용되는 데이터셋 중 하나이다. 그러나 무작위 분할 방식에서는 테스트 픽셀의 상당 부분이 학습 픽셀과 바로 인접하게 놓여 보고되는 정확도를 부풀린다. 본 연구는 공간적 분리를 모델의 수용 영역(receptive field)과 연결하는 누수 없는(leakage-free) 평가 프로토콜을 제안한다. 고전적, 스펙트럴(spectral), 스펙트럴-공간(spectral-spatial), 트랜스포머, vision-backbone, state-space 계열을 포함한 열 가지 서로 다른 아키텍처에 이 프로토콜을 적용한 결과, Macro-F1이 평균 0.147 하락하고 모델 순위가 최대 다섯 계단까지 변한다. 또한 누수 없는 평가는 주어진 벤치마크에서 어떤 아키텍처를 검증할 수 있는지 제한한다. 각 분할(partition)은 유한 반경 내에서만 패치를 지원하므로, 공정한 비교를 위해 수용 영역과 함께 이 반경을 보고하는 것이 필수적이다. 아울러 본 연구는 열 가지 아키텍처가 대체로 동일한 픽셀을 오분류함을 밝혀, 이들 중 어느 것도 해소하지 못하는 데이터 내 스펙트럴 모호성(spectral ambiguity)을 시사한다.
Video smoke detection is a promising fire detection method especially in open or large spaces and outdoor environments. Traditional video smoke detection methods usually consist of candidate region extraction and classification, but lack powerful characterization for smoke. In this paper, we propose a novel video smoke detection method based on deep saliency network. Visual saliency detection aims to highlight the most important object regions in an image. The pixel-level and object-level salient convolutional neural networks are combined to extract the informative smoke saliency map. An end-to-end framework for salient smoke detection and existence prediction of smoke is proposed for application in video smoke detection. The deep feature map is combined with the saliency map to predict the existence of smoke in an image. Initial and augmented dataset are built to measure the performance of frameworks with different design strategies. Qualitative and quantitative analysis at frame-level and pixel-level demonstrate the excellent performance of the ultimate framework.
In this article, we focus on decomposing latent representations in generative adversarial networks or learned feature representations in deep autoencoders into semantically controllable factors in a semisupervised manner, without modifying the original trained models. Particularly, we propose factors' decomposer-entangler network (FDEN) that learns to decompose a latent representation into mutually independent factors. Given a latent representation, the proposed framework draws a set of interpretable factors, each aligned to independent factors of variations by minimizing their total correlation in an information-theoretic means. As a plug-in method, we have applied our proposed FDEN to the existing networks of adversarially learned inference and pioneer network and performed computer vision tasks of image-to-image translation in semantic ways, e.g., changing styles, while keeping the identity of a subject, and object classification in a few-shot learning scheme. We have also validated the effectiveness of the proposed method with various ablation studies in the qualitative, quantitative, and statistical examination.
Hyperspectral image (HSI) denoising is of crucial importance for many subsequent applications, such as HSI classification and interpretation. In this paper, we propose an attention-based deep residual network to directly learn a mapping from noisy HSI to the clean one. To jointly utilize the spatial-spectral information, the current band and its $K$ adjacent bands are simultaneously exploited as the input. Then, we adopt convolution layer with different filter sizes to fuse the multi-scale feature, and use shortcut connection to incorporate the multi-level information for better noise removal. In addition, the channel attention mechanism is employed to make the network concentrate on the most relevant auxiliary information and features that are beneficial to the denoising process best. To ease the training procedure, we reconstruct the output through a residual mode rather than a straightforward prediction. Experimental results demonstrate that our proposed ADRN scheme outperforms the state-of-the-art methods both in quantitative and visual evaluations.
The collection of a high number of pixel-based labeled training samples for tree species identification is time consuming and costly in operational forestry applications. To address this problem, in this paper we investigate the effectiveness of explanation methods for deep neural networks in performing weakly supervised semantic segmentation using only image-level labels. Specifically, we consider four methods:i) class activation maps (CAM); ii) gradient-based CAM; iii) pixel correlation module; and iv) self-enhancing maps (SEM). We compare these methods with each other using both quantitative and qualitative measures of their segmentation accuracy, as well as their computational requirements. Experimental results obtained on an aerial image archive show that:i) considered explanation techniques are highly relevant for the identification of tree species with weak supervision; and ii) the SEM outperforms the other considered methods. The code for this paper is publicly available at https://git.tu-berlin.de/rsim/rs_wsss.
Visual representations underlie object recognition tasks, but they often contain both robust and non-robust features. Our main observation is that image classifiers may perform poorly on out-of-distribution samples because spurious correlations between non-robust features and labels can be changed in a new environment. By analyzing procedures for out-of-distribution generalization with a causal graph, we show that standard classifiers fail because the association between images and labels is not transportable across settings. However, we then show that the causal effect, which severs all sources of confounding, remains invariant across domains. This motivates us to develop an algorithm to estimate the causal effect for image classification, which is transportable (i.e., invariant) across source and target environments. Without observing additional variables, we show that we can derive an estimand for the causal effect under empirical assumptions using representations in deep models as proxies. Theoretical analysis, empirical results, and visualizations show that our approach captures causal invariances and improves overall generalization.
In this paper we introduce the Temporo-Spatial Vision Transformer (TSViT), a fully-attentional model for general Satellite Image Time Series (SITS) processing based on the Vision Transformer (ViT). TSViT splits a SITS record into non-overlapping patches in space and time which are tokenized and subsequently processed by a factorized temporo-spatial encoder. We argue, that in contrast to natural images, a temporal-then-spatial factorization is more intuitive for SITS processing and present experimental evidence for this claim. Additionally, we enhance the model's discriminative power by introducing two novel mechanisms for acquisition-time-specific temporal positional encodings and multiple learnable class tokens. The effect of all novel design choices is evaluated through an extensive ablation study. Our proposed architecture achieves state-of-the-art performance, surpassing previous approaches by a significant margin in three publicly available SITS semantic segmentation and classification datasets. All model, training and evaluation codes are made publicly available to facilitate further research.
뇌의 강건한 지각은 흔히 고차원 집단 활동과 반복되는 구조를 강화하는 국소 가소성 메커니즘의 결합으로 설명된다. 반면 대부분의 현대 이미지 인식 시스템은 오차 역전파와 종단 간 그래디언트 최적화로 학습되는데, 이는 국소 계산 및 국소 가소성과 자연스럽게 부합하지 않는다. 본 논문은 학습되지 않은 리저버 동역학과 자기조직화 헤브 프로토타입 리드아웃을 결합한 강건한 이미지 분류용 생체 모사 학습 전략인 RECAP(Reservoir Computing with Hebbian Co-Activation Prototypes)을 제안한다. RECAP은 시간 평균된 리저버 응답을 활성화 수준으로 이산화하고, 리저버 유닛 쌍에 대한 공동 활성화 마스크를 구성하며, 헤브 방식의 강화-감쇠 규칙으로 클래스별 프로토타입 행렬을 점진적으로 갱신한다. 추론은 중첩 기반 프로토타입 매칭으로 수행된다. 이 방법은 오차 역전파를 사용하지 않으며 온라인 프로토타입 갱신과 자연스럽게 호환된다. 우리는 MNIST-C에서 그 결과로 나타나는 강건성을 보이며, RECAP은 손상된 학습 샘플에 노출되지 않고도 다양한 손상 조건에서 강건성을 유지한다.
Point-image alignment learns useful object representations, but different local feature patterns can produce the same pooled descriptor and receive the same alignment signal. We introduce MS-Point, a cross-modal pretraining framework that directly supervises intermediate point features and their neighborhood responses. At selected encoder stages, we compute neighborhood averages and residuals before pooling and align their branch descriptors with image features. Branch-level objectives align each representation with its visual target, while cross-level consistency and adaptive fusion coordinate information across branches. A common interface of intermediate features and spatial neighborhoods supports both graph encoders and point Transformers. Evaluations across object classification, few-shot adaptation, dense prediction, and low-label scene learning demonstrate the transfer value of this design. Same-backbone comparisons and capacity-matched and computation-matched controls connect these improvements to the structure of supervision. Together, our findings support directly supervising intermediate features and neighborhood variation to learn transferable, geometry-aware representations across modalities.
The book attempts to introduce a gentle introduction to the field of Facial Micro Expressions Recognition (FMER) using Color and Depth images, with the aid of MATLAB programming environment. FMER is a subset of image processing and it is a multidisciplinary topic to analysis. So, it requires familiarity with other topics of Artifactual Intelligence (AI) such as machine learning, digital image processing, psychology and more. So, it is a great opportunity to write a book which covers all of these topics for beginner to professional readers in the field of AI and even without having background of AI. Our goal is to provide a standalone introduction in the field of MFER analysis in the form of theorical descriptions for readers with no background in image processing with reproducible Matlab practical examples. Also, we describe any basic definitions for FMER analysis and MATLAB library which is used in the text, that helps final reader to apply the experiments in the real-world applications. We believe that this book is suitable for students, researchers, and professionals alike, who need to develop practical skills, along with a basic understanding of the field. We expect that, after reading this book, the reader feels comfortable with different key stages such as color and depth image processing, color and depth image representation, classification, machine learning, facial micro-expressions recognition, feature extraction and dimensionality reduction. The book attempts to introduce a gentle introduction to the field of Facial Micro Expressions Recognition (FMER) using Color and Depth images, with the aid of MATLAB programming environment.
Automated detection of Gallbladder Cancer (GBC) from Ultrasound (US) images is an important problem, which has drawn increased interest from researchers. However, most of these works use difficult-to-acquire information such as bounding box annotations or additional US videos. In this paper, we focus on GBC detection using only image-level labels. Such annotation is usually available based on the diagnostic report of a patient, and do not require additional annotation effort from the physicians. However, our analysis reveals that it is difficult to train a standard image classification model for GBC detection. This is due to the low inter-class variance (a malignant region usually occupies only a small portion of a US image), high intra-class variance (due to the US sensor capturing a 2D slice of a 3D object leading to large viewpoint variations), and low training data availability. We posit that even when we have only the image level label, still formulating the problem as object detection (with bounding box output) helps a deep neural network (DNN) model focus on the relevant region of interest. Since no bounding box annotations is available for training, we pose the problem as weakly supervised object detection (WSOD). Motivated by the recent success of transformer models in object detection, we train one such model, DETR, using multi-instance-learning (MIL) with self-supervised instance selection to suit the WSOD task. Our proposed method demonstrates an improvement of AP and detection sensitivity over the SOTA transformer-based and CNN-based WSOD methods. Project page is at https://gbc-iitd.github.io/wsod-gbc
We present an efficient 3D object detection framework based on a single RGB image in the scenario of autonomous driving. Our efforts are put on extracting the underlying 3D information in a 2D image and determining the accurate 3D bounding box of the object without point cloud or stereo data. Leveraging the off-the-shelf 2D object detector, we propose an artful approach to efficiently obtain a coarse cuboid for each predicted 2D box. The coarse cuboid has enough accuracy to guide us to determine the 3D box of the object by refinement. In contrast to previous state-of-the-art methods that only use the features extracted from the 2D bounding box for box refinement, we explore the 3D structure information of the object by employing the visual features of visible surfaces. The new features from surfaces are utilized to eliminate the problem of representation ambiguity brought by only using a 2D bounding box. Moreover, we investigate different methods of 3D box refinement and discover that a classification formulation with quality aware loss has much better performance than regression. Evaluated on the KITTI benchmark, our approach outperforms current state-of-the-art methods for single RGB image based 3D object detection.
To enable a deep learning-based system to be used in the medical domain as a computer-aided diagnosis system, it is essential to not only classify diseases but also present the locations of the diseases. However, collecting instance-level annotations for various thoracic diseases is expensive. Therefore, weakly supervised localization methods have been proposed that use only image-level annotation. While the previous methods presented the disease location as the most discriminative part for classification, this causes a deep network to localize wrong areas for indistinguishable X-ray images. To solve this issue, we propose a spatial attention method using disease masks that describe the areas where diseases mainly occur. We then apply the spatial attention to find the precise disease area by highlighting the highest probability of disease occurrence. Meanwhile, the various sizes, rotations and noise in chest X-ray images make generating the disease masks challenging. To reduce the variation among images, we employ an alignment module to transform an input X-ray image into a generalized image. Through extensive experiments on the NIH-Chest X-ray dataset with eight kinds of diseases, we show that the proposed method results in superior localization performances compared to state-of-the-art methods.
Knowledge of search-landscape features of BlackBox Optimization (BBO) problems offers valuable information in light of the Algorithm Selection and/or Configuration problems. Exploratory Landscape Analysis (ELA) models have gained success in identifying predefined human-derived features and in facilitating portfolio selectors to address those challenges. Unlike ELA approaches, the current study proposes to transform the identification problem into an image recognition problem, with a potential to detect conception-free, machine-driven landscape features. To this end, we introduce the notion of Landscape Images, which enables us to generate imagery instances per a benchmark function, and then target the classification challenge over a diverse generalized dataset of functions. We address it as a supervised multi-class image recognition problem and apply basic artificial neural network models to solve it. The efficacy of our approach is numerically validated on the noise free BBOB and IOHprofiler benchmarking suites. This evident successful learning is another step toward automated feature extraction and local structure deduction of BBO problems. By using this definition of landscape images, and by capitalizing on existing capabilities of image recognition algorithms, we foresee the construction of an ImageNet-like library of functions for training generalized detectors that rely on machine-driven features.