Few-shot Weakly-Supervised Object Detection via Directional Statistics

For alleviating the human labor associated with annotating the training data for learning object detectors, recent research has focused on semi-supervised object detection (SSOD) and weakly supervised object detection (WSOD) approaches. In SSOD, instead of annotating all the instances in the whole training set, people only need to annotate the part of the training instances using bounding boxes. In WSOD, people need to annotate the image-level tags on all training images to indicate the object categories contained by the corresponding images since more detailed bounding box annotations are no longer needed. Along this line of research, this paper makes a further step to alleviate the human labor in annotating training data, leading to the problem of object detection with semi-annotated weak labels (ODSAWLs). Instead of labeling image-level tags on all training images, ODSAWL only needs the image-level tags for a small portion of the training images, and then, the object detectors can be learned from a small portion of the weakly-labeled training images and from the remaining unlabeled training images. To address such a challenging problem, this paper proposes a cross model co-training framework that collaborates an object localizer and a tag generator in an alternative optimization procedure. Specifically, during the learning procedure, these two (deep) models can transfer the needed knowledge (including labels and visual patterns) between each other. The whole learning procedure is accomplished in a few stages under the guidance of a progressive learning curriculum. To demonstrate the effectiveness of the proposed approach, we implement the comprehensive experiments on three benchmark datasets, where the obtained experimental results are quite encouraging. Notably, by using only about 15% weakly labeled training images, the proposed approach can effectively approach, or even outperform, the state-of-the-art WSOD methods.

Deep Learning for Weakly-Supervised Object Detection and Localization: A Survey

Weakly supervised object detection (WSOD) in remote sensing images (RSI) only require image-level labels to detect various objects. Most of the WSOD methods incline to capture the most discriminative parts of object rather than the entire object, and the number of easy and hard samples is imbalanced. To address the first problem, a novel metric named objectness score (OS) is proposed and incorporated into the training loss of our WSOD model. The OS is consisted of the traditional class confidence score (CCS) and the object completeness prior score (OCPS). The CCS can provide the probability that a proposal belongs to a certain class, and the OCPS can quantify the completeness that a proposal covers the entire object. Therefore, the samples which cover the entire object with high class confidences will be assigned large weight in the training loss through OS. To handle the second problem, a novel metric named difficulty evaluation score (DES) is proposed and also incorporated into the training loss. The DES is calculated by using the entropy of confidence score vector of each proposal and is used to quantify how difficult a proposal can be identified correctly, consequently, the hard samples will also be assigned large weight in the training loss through DES. The ablation experiments on two RSI datasets verify the effectiveness of the proposed OS and DES. The comprehensive quantitative and subjective evaluations demonstrate that our method inclines to detect the entire object accurately, and surpasses seven state-of-the-art WSOD methods.

Weakly supervised object detection (WSOD) in remote sensing images (RSIs) has attracted lots of attention because it solely employs image-level labels to drive the model training. Most of the WSOD methods incline to mine salient object as positive instance, and the less salient objects are considered as negative instances, which will cause the problem of missing instances. In addition, the quantity of hard and easy instances is usually imbalanced, consequently, the cumulative loss of a large amount of easy instances dominates the training loss, which limits the upper bound of WSOD performance. To handle the first problem, a complementary detection network (CDN) is proposed, which consists of a complementary multiple instance detection network (CMIDN) and a complementary feature learning (CFL) module. The CDN can capture robust complementary information from two basic multiple instance detection networks (MIDNs) and mine more object instances. To handle the second problem, an instance difficulty evaluation metric named instance difficulty score (IDS) is proposed, which is employed as the weight of each instance in the training loss. Consequently, the hard instances will be assigned larger weights according to the IDS, which can improve the upper bound of WSOD performance. The ablation experiments demonstrate that our method significantly increases the baseline method by large margins, i.e. 23.6% (10.2%) mAP and 32.4% (13.1%) CorLoc gains on the NWPU VHR-10.v2 (DIOR) dataset. Our method obtains 58.1% (26.7%) mAP and 72.4% (47.9%) CorLoc on the NWPU VHR-10.v2 (DIOR) dataset, which achieves better performance compared with seven advanced WSOD methods.

A weakly supervised object detection (WSOD) task is to train a detector with only image-level labels provided. Except for the training difficulty introduced by weaker annotations, the inherent complexity of the remote-sensing images (RSIs) also adds to the challenge. To boost the detector’s localization accuracy, we aim to exploit more semantic information contained in images and help improve the general robustness of the model. Noticing previous methods tend to focus on the most discriminative part of an object, we design a self-training-based network that leverages local semantic features. To this end, we develop a semantic-balanced localization module (SBLM) that distinguishes foreground from background and accurate proposals from incomplete ones, by leveraging a balance of region of interest (ROI) and its context information. Moreover, we find that the self-training strategy highly relies on the quality of pseudo-ground-truth boxes. Motivated by this possible lack of robustness, we design a comprehensive clustering module (CCM) and saliency-based proposal filtering (SPF) module that select pseudo-ground truth more comprehensively under supervision. To be more specific, CCM aims to reduce the arbitrariness during assigning pseudo-labels by considering multiple categorical vectors simultaneously. Salient object detection (SOD) is applied in the SPF module to help evaluate the quality of the chosen pseudo-ground-truth boxes. The detection performance is significantly boosted with the proposed method. Extensive experiments conducted on the NWPU VHR-10.v2 dataset and the DIOR dataset validate that the proposed model outperforms the previous state-of-the-art methods favorably with an mAP of 64.9% and 28.1%, respectively.

A generic approach for the extraction and recognition of gesture using raw grey-level images is presented. The probabilistic visual learning approach, a learning method recently proposed by B. Moghaddam and A. Pentland (1997), is used to create a set of compact statistical representations of gesture appearance on low dimensional eigenspaces. The same probabilistic modeling framework is used to extract and track gesture and to perform gesture recognition over long image sequences. Gesture extraction and tracking are based on maximum likelihood gesture detection in the input image. Recognition is performed by using the set of learned probabilistic appearance models as estimates of the emission probabilities of a continuous density hidden Markov model (CDHMM). Although the segmentation and CDHMM-based recognition use raw grey-level images, the method is fast, thanks to the data compression obtained by probabilistic visual learning. The approach is comprehensive and may be applied to other visual motion recognition tasks. It does not require application-tailored extraction of image features, the use of markers or gloves. A real-time implementation of the method on a standard PC-based vision system is under consideration.

High dimensional directional data is becoming increasingly important in contemporary applications such as analysis of text and gene-expression data. A natural model for multi-variate directional data is provided by the von Mises-Fisher (vMF) distribution on the unit hypersphere that is analogous to the multi-variate Gaussian distribution in Rd. In this paper, we propose modeling complex directional data as a mixture of vMF distributions. We derive and analyze two variants of the Expectation Maximization (EM) framework for estimating the parameters of this mixture. We also propose two clustering algorithms corresponding to these variants. An interesting aspect of our methodology is that the spherical kmeans algorithm (kmeans with cosine similarity) can be shown to be a special case of both our algorithms. Thus, modeling text data by vMF distributions lends theoretical validity to the use of cosine similarity which has been widely used by the information retrieval community. As part of experimental validation, we present results on modeling high-dimensional text and gene-expression data as a mixture of vMF distributions. The results indicate that our approach yields superior clusterings especially for difficult clustering tasks in high-dimensional spaces.

We consider the problem of semantic segmentation, i.e. assigning each pixel in an image to a set of pre-defined semantic object categories. State-of-the-art semantic segmentation algorithms typically consist of three components: a local appearance model, a local consistency model and a global consistency model. These three components are generally integrated into a unified probabilistic framework. While it enables at training time a joint estimation of the model parameters and while it ensures at test time a globally consistent labeling of the pixels, it also comes at a high computational cost. We propose a simple approach to semantic segmentation where the three components are decoupled (this journal submission is an extended version of the following conference paper: G. Csurka and F. Perronnin, "A simple high performance approach to semantic segmentation", BMVC, 2008). For the local appearance model, we make use of the Fisher kernel. While this framework was shown to lead to high accuracy for image classification, to our best knowledge this is its first application to the segmentation problem. The semantic segmentation process is then guided by a low-level segmentation which enforces local consistency. Finally, to enforce image-level consistency we use global image classifiers: if an image as a whole is unlikely to contain an object class, then the corresponding class is not considered in the segmentation pipeline. The decoupling of the components makes our system very efficient both at training and test time. An efficient training enables to estimate the model parameters on large quantities of data. Especially, we explain how our system can leverage weakly labeled data, i.e. images for which we do not have pixel-level labels but either object bounding boxes or even only image-level labels. We believe that an important contribution of this paper is to show that even a simple decoupled system can provide state-of-the-art performance on the PASCAL VOC 2007, PASCAL VOC 2008 and MSRC 21 datasets.

Capturing word meaning is one of the challenges of natural language processing (NLP). Formal models of meaning, such as networks of words or concepts, are knowledge repositories used in a variety of applications. To be effectively used, these networks have to be large or, at least, adapted to specific domains. Learning word meaning from texts is then an active area of research. Lexico-syntactic pattern methods are one of the possible solutions. Yet, these models do not use structural properties of target semantic relations, e.g. transitivity, during learning. In this paper, we propose a novel lexico-syntactic pattern probabilistic method for learning taxonomies that explicitly models transitivity and naturally exploits vector space model techniques for reducing space dimensions. We define two probabilistic models: the direct probabilistic model and the induced probabilistic model. The first is directly estimated on observations over text collections. The second uses transitivity on the direct probabilistic model to induce probabilities of derived events. Within our probabilistic model, we also propose a novel way of using singular value decomposition as unsupervised method for feature selection in estimating direct probabilities. We empirically show that the induced probabilistic taxonomy learning model outperforms state-of-the-art probabilistic models and our unsupervised feature selection method improves performance.

Automatic prostate segmentation from MR images is an important task in various clinical applications such as prostate cancer staging and MR-guided radiotherapy planning. However, the large appearance and shape variations of the prostate in MR images make the segmentation problem difficult to solve. Traditional Active Shape/Appearance Model (ASM/AAM) has limited accuracy on this problem, since its basic assumption, i.e., both shape and appearance of the targeted organ follow Gaussian distributions, is invalid in prostate MR images. To this end, the authors propose a sparse dictionary learning method to model the image appearance in a nonparametric fashion and further integrate the appearance model into a deformable segmentation framework for prostate MR segmentation. To drive the deformable model for prostate segmentation, the authors propose nonparametric appearance and shape models. The nonparametric appearance model is based on a novel dictionary learning method, namely distributed discriminative dictionary (DDD) learning, which is able to capture fine distinctions in image appearance. To increase the differential power of traditional dictionary-based classification methods, the authors' DDD learning approach takes three strategies. First, two dictionaries for prostate and nonprostate tissues are built, respectively, using the discriminative features obtained from minimum redundancy maximum relevance feature selection. Second, linear discriminant analysis is employed as a linear classifier to boost the optimal separation between prostate and nonprostate tissues, based on the representation residuals from sparse representation. Third, to enhance the robustness of the authors' classification method, multiple local dictionaries are learned for local regions along the prostate boundary (each with small appearance variations), instead of learning one global classifier for the entire prostate. These discriminative dictionaries are located on different patches of the prostate surface and trained to adaptively capture the appearance in different prostate zones, thus achieving better local tissue differentiation. For each local region, multiple classifiers are trained based on the randomly selected samples and finally assembled by a specific fusion method. In addition to this nonparametric appearance model, a prostate shape model is learned from the shape statistics using a novel approach, sparse shape composition, which can model nonGaussian distributions of shape variation and regularize the 3D mesh deformation by constraining it within the observed shape subspace. The proposed method has been evaluated on two datasets consisting of T2-weighted MR prostate images. For the first (internal) dataset, the classification effectiveness of the authors' improved dictionary learning has been validated by comparing it with three other variants of traditional dictionary learning methods. The experimental results show that the authors' method yields a Dice Ratio of 89.1% compared to the manual segmentation, which is more accurate than the three state-of-the-art MR prostate segmentation methods under comparison. For the second dataset, the MICCAI 2012 challenge dataset, the authors' proposed method yields a Dice Ratio of 87.4%, which also achieves better segmentation accuracy than other methods under comparison. A new magnetic resonance image prostate segmentation method is proposed based on the combination of deformable model and dictionary learning methods, which achieves more accurate segmentation performance on prostate T2 MR images.

Entropy-based closure for probabilistic learning on manifolds

Due to advances in technology, there is a presence of directional data in a wide variety of fields. Often distributions to model directional data are defined on manifold or constrained spaces. Regular statistical methods applied to data defined on special geometries can give misleading results, and this demands new statistical theory. This dissertation addresses two such problems and develops Bayesian methodologies to improve inference in these arenas. It consists of two projects: 1. A Bayesian Methodology for Estimation for Sparse Canonical Correlation, and 2. Bayesian Analysis of Finite Mixture Model for Spherical Data. In principle, it can be challenging to integrate data measured on the same individuals occurring from different experiments and model it together to gain a larger understanding of the problem. Canonical Correlation Analysis (CCA) provides a useful tool for establishing relationships between such data sets. When dealing with high dimensional data sets, Structured Sparse CCA (ScSCCA) is a rapidly developing methodological area which seeks to represent the interrelations using sparse direction vectors for CCA. There is less development in Bayesian methodology in this area. We propose a novel Bayesian ScSCCA method with the use of a Bayesian infinite factor model. Using a multiplicative half Cauchy prior process, we bring in sparsity at the level of the projection matrix. Additionally, we promote further sparsity in the covariance matrix by using graphical horseshoe prior or diagonal structure. We compare the results for our proposed model with competing frequentist and Bayesian methods and apply the developed method to omics data arising from a breast cancer study. In the second project, we perform Bayesian Analysis for the von Mises Fisher (vMF) distribution on the sphere which is a common and important distribution used for directional data. In the first part of this project, we propose a new conjugate prior for the mean vector and concentration parameter of the vMF distribution. Further we prove its properties like finiteness, unimodality, and provide interpretations of its hyperparameters. In the second part, we utilize a popular prior structure for a mixture of vMF distributions. In this case, the posterior of the concentration parameter consists of an intractable Bessel function of the first kind. We propose a novel Data Augmentation Strategy (DAS) using a Negative Binomial Distribution that removes this intractable Bessel function. Furthermore, we apply the developed methodology to Diffusion Tensor Imaging (DTI) data for clustering to explore voxel connectivity in human brain.

The geographical identification of content in Social Networks have enabled to bridge the gap between online social platforms and the physical world. Although vast amounts of data in such networks are due to breaking news or global occurrences, local events witnessed by users in situ are also present in these streams and of great importance for many city entities. Nowadays, unsupervised machine learning techniques, such as Tweet-SCAN, are able to retrospectively detect these local events from tweets. However, these approaches have limited abilities to reason about unseen observations in a principled way due to the lack of a proper probabilistic foundation. Probabilistic models have also been proposed for the task, but their event identification capabilities are far from those of Tweet-SCAN. In this paper, we identify two key factors which, when combined, boost the accuracy of such models. As a first key factor, we notice that the large amount of meaningless social data requires explicitly modeling non-event observations.Therefore, we propose to incorporate a background model that captures spatio-temporal fluctuations of non-event tweets. As a second key factor, we observe that the shortness of tweets hampers the application of traditional topic models. Thus, we integrate event detection and topic modeling, assigning topic proportions to events instead of assigning them to individual tweets. As a result, we propose Warble, a new probabilistic model and learning scheme for retrospective event detection that incorporates these two key factors. We evaluate Warble in a data set of tweets located in Barcelona during its festivities. The empirical results show that the model outperforms other state-of-the-art techniques in detecting various types of events while relying on a principled probabilistic framework that enables to reason under uncertainty.

Training object detectors from few weakly-labeled and many unlabeled images

With the increasing integration of AI in medical research and applications, the issue of fairness has become as critical as diagnostic accuracy. In dermatology diagnosis, the challenge of class-imbalanced data, which is sometimes limited and contains demographic attributes, results in an imbalanced and insufficient representation within the feature space of deep learning models. Besides, feature entanglement within deep learning models confuses skin tone and disease condition information, impairing model performance among vulnerable groups. Moreover, feature entanglement often constrains the efforts to mitigate unfairness, entailing a trade-off between fairness and diagnostic accuracy. This paper introduces the Disentangled Feature Contrastive learning and Augmentation framework (DFCA), aiming to enhance fairness in dermatological diagnoses without compromising accuracy. Initially, DFCA disentangles skin images into disease related and skin-tone features. Subsequently, the two sets of features are projected into normalized spaces for contrastive learning, each modeled by a mixture of von Mises-Fisher (vMF) distributions. DFCA then samples from these vMF distributions to inversely augment the feature space. To further evaluate the fairness-accuracy balance, we propose a new metric, the Accuracy-Fairness Balance Degree (AFBD). Extensive experiments demonstrate that DFCA significantly improves both fairness and accuracy compared to state-of-the-art methods.