Lack of interpretability of deep convolutional neural networks (DCNN) is a well-known problem particularly in the medical domain as clinicians want trustworthy automated decisions. One way to improve trust is to demonstrate the localisation of feature representations with respect to expert labeled regions of interest. In this work, we investigate the localisation of features learned via two varied learning paradigms and demonstrate the superiority of one learning approach with respect to localisation. Our analysis on medical and natural datasets show that the traditional end-to-end (E2E) learning strategy has a limited ability to localise discriminative features across multiple network layers. We show that a layer-wise learning strategy, namely cascade learning (CL), results in more localised features. Considering localisation accuracy, we not only show that CL outperforms E2E but that it is a promising method of predicting regions. On the YOLO object detection framework, our best result shows that CL outperforms the E2E scheme by $2\%$ in mAP.
The foundation model has recently garnered significant attention due to its potential to revolutionize the field of visual representation learning in a self-supervised manner. While most foundation models are tailored to effectively process RGB images for various visual tasks, there is a noticeable gap in research focused on spectral data, which offers valuable information for scene understanding, especially in remote sensing (RS) applications. To fill this gap, we created for the first time a universal RS foundation model, named SpectralGPT, which is purpose-built to handle spectral RS images using a novel 3D generative pretrained transformer (GPT). Compared to existing foundation models, SpectralGPT 1) accommodates input images with varying sizes, resolutions, time series, and regions in a progressive training fashion, enabling full utilization of extensive RS big data; 2) leverages 3D token generation for spatial-spectral coupling; 3) captures spectrally sequential patterns via multi-target reconstruction; 4) trains on one million spectral RS images, yielding models with over 600 million parameters. Our evaluation highlights significant performance improvements with pretrained SpectralGPT models, signifying substantial potential in advancing spectral RS big data applications within the field of geoscience across four downstream tasks: single/multi-label scene classification, semantic segmentation, and change detection.
The recent development of deep learning methods applied to vision has enabled their increasing integration into real-world applications to perform complex Computer Vision (CV) tasks. However, image acquisition conditions have a major impact on the performance of high-level image processing. A possible solution to overcome these limitations is to artificially augment the training databases or to design deep learning models that are robust to signal distortions. We opt here for the first solution by enriching the database with complex and realistic distortions which were ignored until now in the existing databases. To this end, we built a new versatile database derived from the well-known MS-COCO database to which we applied local and global photo-realistic distortions. These new local distortions are generated by considering the scene context of the images that guarantees a high level of photo-realism. Distortions are generated by exploiting the depth information of the objects in the scene as well as their semantics. This guarantees a high level of photo-realism and allows to explore real scenarios ignored in conventional databases dedicated to various CV applications. Our versatile database offers an efficient solution to improve the robustness of various CV tasks such as Object Detection (OD), scene segmentation, and distortion-type classification methods. The image database, scene classification index, and distortion generation codes are publicly available \footnote{\url{https://github.com/Aymanbegh/CD-COCO}}
Recent advancements have significantly improved the efficiency and effectiveness of deep learning methods for imagebased remote sensing tasks. However, the requirement for large amounts of labeled data can limit the applicability of deep neural networks to existing remote sensing datasets. To overcome this challenge, fewshot learning has emerged as a valuable approach for enabling learning with limited data. While previous research has evaluated the effectiveness of fewshot learning methods on satellite based datasets, little attention has been paid to exploring the applications of these methods to datasets obtained from UAVs, which are increasingly used in remote sensing studies. In this review, we provide an up to date overview of both existing and newly proposed fewshot classification techniques, along with appropriate datasets that are used for both satellite based and UAV based data. Our systematic approach demonstrates that fewshot learning can effectively adapt to the broader and more diverse perspectives that UAVbased platforms can provide. We also evaluate some SOTA fewshot approaches on a UAV disaster scene classification dataset, yielding promising results. We emphasize the importance of integrating XAI techniques like attention maps and prototype analysis to increase the transparency, accountability, and trustworthiness of fewshot models for remote sensing. Key challenges and future research directions are identified, including tailored fewshot methods for UAVs, extending to unseen tasks like segmentation, and developing optimized XAI techniques suited for fewshot remote sensing problems. This review aims to provide researchers and practitioners with an improved understanding of fewshot learnings capabilities and limitations in remote sensing, while highlighting open problems to guide future progress in efficient, reliable, and interpretable fewshot methods.
Most deep learning-based acoustic scene classification (ASC) approaches identify scenes based on acoustic features converted from audio clips containing mixed information entangled by polyphonic audio events (AEs). However, these approaches have difficulties in explaining what cues they use to identify scenes. This paper conducts the first study on disclosing the relationship between real-life acoustic scenes and semantic embeddings from the most relevant AEs. Specifically, we propose an event-relational graph representation learning (ERGL) framework for ASC to classify scenes, and simultaneously answer clearly and straightly which cues are used in classifying. In the event-relational graph, embeddings of each event are treated as nodes, while relationship cues derived from each pair of nodes are described by multi-dimensional edge features. Experiments on a real-life ASC dataset show that the proposed ERGL achieves competitive performance on ASC by learning embeddings of only a limited number of AEs. The results show the feasibility of recognizing diverse acoustic scenes based on the audio event-relational graph. Visualizations of graph representations learned by ERGL are available here (https://github.com/Yuanbo2020/ERGL).
The correlation between the sharpness of loss minima and generalisation in the context of deep neural networks has been subject to discussion for a long time. Whilst mostly investigated in the context of selected benchmark data sets in the area of computer vision, we explore this aspect for the acoustic scene classification task of the DCASE2020 challenge data. Our analysis is based on two-dimensional filter-normalised visualisations and a derived sharpness measure. Our exploratory analysis shows that sharper minima tend to show better generalisation than flat minima -even more so for out-of-domain data, recorded from previously unseen devices-, thus adding to the dispute about better generalisation capabilities of flat minima. We further find that, in particular, the choice of optimisers is a main driver of the sharpness of minima and we discuss resulting limitations with respect to comparability. Our code, trained model states and loss landscape visualisations are publicly available.
We present a locality-aware method for interpreting the latent space of wavelet-based Generative Adversarial Networks (GANs), that can well capture the large spatial and spectral variability that is characteristic to satellite imagery. By focusing on preserving locality, the proposed method is able to decompose the weight-space of pre-trained GANs and recover interpretable directions that correspond to high-level semantic concepts (such as urbanization, structure density, flora presence) - that can subsequently be used for guided synthesis of satellite imagery. In contrast to typically used approaches that focus on capturing the variability of the weight-space in a reduced dimensionality space (i.e., based on Principal Component Analysis, PCA), we show that preserving locality leads to vectors with different angles, that are more robust to artifacts and can better preserve class information. Via a set of quantitative and qualitative examples, we further show that the proposed approach can outperform both baseline geometric augmentations, as well as global, PCA-based approaches for data synthesis in the context of data augmentation for satellite scene classification.
Vision Transformer (ViT) models have recently emerged as powerful and versatile models for various visual tasks. Recently, a work called PMF has achieved promising results in few-shot image classification by utilizing pre-trained vision transformer models. However, PMF employs full fine-tuning for learning the downstream tasks, leading to significant overfitting and storage issues, especially in the remote sensing domain. In order to tackle these issues, we turn to the recently proposed parameter-efficient tuning methods, such as VPT, which updates only the newly added prompt parameters while keeping the pre-trained backbone frozen. Inspired by VPT, we propose the Meta Visual Prompt Tuning (MVP) method. Specifically, we integrate the VPT method into the meta-learning framework and tailor it to the remote sensing domain, resulting in an efficient framework for Few-Shot Remote Sensing Scene Classification (FS-RSSC). Furthermore, we introduce a novel data augmentation strategy based on patch embedding recombination to enhance the representation and diversity of scenes for classification purposes. Experiment results on the FS-RSSC benchmark demonstrate the superior performance of the proposed MVP over existing methods in various settings, such as various-way-various-shot, various-way-one-shot, and cross-domain adaptation.
The increasing availability of multi-sensor data sparks wide interest in multimodal self-supervised learning. However, most existing approaches learn only common representations across modalities while ignoring intra-modal training and modality-unique representations. We propose Decoupling Common and Unique Representations (DeCUR), a simple yet effective method for multimodal self-supervised learning. By distinguishing inter- and intra-modal embeddings through multimodal redundancy reduction, DeCUR can integrate complementary information across different modalities. We evaluate DeCUR in three common multimodal scenarios (radar-optical, RGB-elevation, and RGB-depth), and demonstrate its consistent improvement regardless of architectures and for both multimodal and modality-missing settings. With thorough experiments and comprehensive analysis, we hope this work can provide valuable insights and raise more interest in researching the hidden relationships of multimodal representations.
Visual Saliency refers to the innate human mechanism of focusing on and extracting important features from the observed environment. Recently, there has been a notable surge of interest in the field of automotive research regarding the estimation of visual saliency. While operating a vehicle, drivers naturally direct their attention towards specific objects, employing brain-driven saliency mechanisms that prioritize certain elements over others. In this investigation, we present an intelligent system that combines a drowsiness detection system for drivers with a scene comprehension pipeline based on saliency. To achieve this, we have implemented a specialized 3D deep network for semantic segmentation, which has been pretrained and tailored for processing the frames captured by an automotive-grade external camera. The proposed pipeline was hosted on an embedded platform utilizing the STA1295 core, featuring ARM A7 dual-cores, and embeds an hardware accelerator. Additionally, we employ an innovative biosensor embedded on the car steering wheel to monitor the driver drowsiness, gathering the PhotoPlethysmoGraphy (PPG) signal of the driver. A dedicated 1D temporal deep convolutional network has been devised to classify the collected PPG time-series, enabling us to assess the driver level of attentiveness. Ultimately, we compare the determined attention level of the driver with the corresponding saliency-based scene classification to evaluate the overall safety level. The efficacy of the proposed pipeline has been validated through extensive experimental results.
Deep neural networks (DNNs) have risen to prominence as key solutions in numerous AI applications for earth observation (AI4EO). However, their susceptibility to adversarial examples poses a critical challenge, compromising the reliability of AI4EO algorithms. This paper presents a novel Universal Adversarial Defense approach in Remote Sensing Imagery (UAD-RS), leveraging pre-trained diffusion models to protect DNNs against universal adversarial examples exhibiting heterogeneous patterns. Specifically, a universal adversarial purification framework is developed utilizing pre-trained diffusion models to mitigate adversarial perturbations through the introduction of Gaussian noise and subsequent purification of the perturbations from adversarial examples. Additionally, an Adaptive Noise Level Selection (ANLS) mechanism is introduced to determine the optimal noise level for the purification framework with a task-guided Frechet Inception Distance (FID) ranking strategy, thereby enhancing purification performance. Consequently, only a single pre-trained diffusion model is required for purifying universal adversarial samples with heterogeneous patterns across each dataset, significantly reducing training efforts for multiple attack settings while maintaining high performance without prior knowledge of adversarial perturbations. Experimental results on four heterogeneous RS datasets, focusing on scene classification and semantic segmentation, demonstrate that UAD-RS outperforms state-of-the-art adversarial purification approaches, providing universal defense against seven commonly encountered adversarial perturbations. Codes and the pre-trained models are available online (https://github.com/EricYu97/UAD-RS).
Deep Learning models like Convolutional Neural Networks (CNN) are powerful image classifiers, but what factors determine whether they attend to similar image areas as humans do? While previous studies have focused on technological factors, little is known about the role of factors that affect human attention. In the present study, we investigated how the tasks used to elicit human attention maps interact with image characteristics in modulating the similarity between humans and CNN. We varied the intentionality of human tasks, ranging from spontaneous gaze during categorization over intentional gaze-pointing up to manual area selection. Moreover, we varied the type of image to be categorized, using either singular, salient objects, indoor scenes consisting of object arrangements, or landscapes without distinct objects defining the category. The human attention maps generated in this way were compared to the CNN attention maps revealed by explainable artificial intelligence (Grad-CAM). The influence of human tasks strongly depended on image type: For objects, human manual selection produced maps that were most similar to CNN, while the specific eye movement task has little impact. For indoor scenes, spontaneous gaze produced the least similarity, while for landscapes, similarity was equally low across all human tasks. To better understand these results, we also compared the different human attention maps to each other. Our results highlight the importance of taking human factors into account when comparing the attention of humans and CNN.
Scene recognition based on deep-learning has made significant progress, but there are still limitations in its performance due to challenges posed by inter-class similarities and intra-class dissimilarities. Furthermore, prior research has primarily focused on improving classification accuracy, yet it has given less attention to achieving interpretable, precise scene classification. Therefore, we are motivated to propose EnTri, an ensemble scene recognition framework that employs ensemble learning using a hierarchy of visual features. EnTri represents features at three distinct levels of detail: pixel-level, semantic segmentation-level, and object class and frequency level. By incorporating distinct feature encoding schemes of differing complexity and leveraging ensemble strategies, our approach aims to improve classification accuracy while enhancing transparency and interpretability via visual and textual explanations. To achieve interpretability, we devised an extension algorithm that generates both visual and textual explanations highlighting various properties of a given scene that contribute to the final prediction of its category. This includes information about objects, statistics, spatial layout, and textural details. Through experiments on benchmark scene classification datasets, EnTri has demonstrated superiority in terms of recognition accuracy, achieving competitive performance compared to state-of-the-art approaches, with an accuracy of 87.69%, 75.56%, and 99.17% on the MIT67, SUN397, and UIUC8 datasets, respectively.
The field of Explainable Artificial Intelligence (XAI) aims to improve the interpretability of black-box machine learning models. Building a heatmap based on the importance value of input features is a popular method for explaining the underlying functions of such models in producing their predictions. Heatmaps are almost understandable to humans, yet they are not without flaws. Non-expert users, for example, may not fully understand the logic of heatmaps (the logic in which relevant pixels to the model's prediction are highlighted with different intensities or colors). Additionally, objects and regions of the input image that are relevant to the model prediction are frequently not entirely differentiated by heatmaps. In this paper, we propose a framework called TbExplain that employs XAI techniques and a pre-trained object detector to present text-based explanations of scene classification models. Moreover, TbExplain incorporates a novel method to correct predictions and textually explain them based on the statistics of objects in the input image when the initial prediction is unreliable. To assess the trustworthiness and validity of the text-based explanations, we conducted a qualitative experiment, and the findings indicated that these explanations are sufficiently reliable. Furthermore, our quantitative and qualitative experiments on TbExplain with scene classification datasets reveal an improvement in classification accuracy over ResNet variants.
We investigate the utility of in-domain self-supervised pre-training of vision models in the analysis of remote sensing imagery. Self-supervised learning (SSL) has emerged as a promising approach for remote sensing image classification due to its ability to exploit large amounts of unlabeled data. Unlike traditional supervised learning, SSL aims to learn representations of data without the need for explicit labels. This is achieved by formulating auxiliary tasks that can be used for pre-training models before fine-tuning them on a given downstream task. A common approach in practice to SSL pre-training is utilizing standard pre-training datasets, such as ImageNet. While relevant, such a general approach can have a sub-optimal influence on the downstream performance of models, especially on tasks from challenging domains such as remote sensing. In this paper, we analyze the effectiveness of SSL pre-training by employing the iBOT framework coupled with Vision transformers trained on Million-AID, a large and unlabeled remote sensing dataset. We present a comprehensive study of different self-supervised pre-training strategies and evaluate their effect across 14 downstream datasets with diverse properties. Our results demonstrate that leveraging large in-domain datasets for self-supervised pre-training consistently leads to improved predictive downstream performance, compared to the standard approaches found in practice.
Varying conditions between the data seen at training and at application time remain a major challenge for machine learning. We study this problem in the context of Acoustic Scene Classification (ASC) with mismatching recording devices. Previous works successfully employed frequency-wise normalization of inputs and hidden layer activations in convolutional neural networks to reduce the recording device discrepancy. The main objective of this work was to adopt frequency-wise normalization for Audio Spectrogram Transformers (ASTs), which have recently become the dominant model architecture in ASC. To this end, we first investigate how recording device characteristics are encoded in the hidden layer activations of ASTs. We find that recording device information is initially encoded in the frequency dimension; however, after the first self-attention block, it is largely transformed into the token dimension. Based on this observation, we conjecture that suppressing recording device characteristics in the input spectrogram is the most effective. We propose a frequency-centering operation for spectrograms that improves the ASC performance on unseen recording devices on average by up to 18.2 percentage points.
Domain shift is considered a challenge in machine learning as it causes significant degradation of model performance. In the Acoustic Scene Classification task (ASC), domain shift is mainly caused by different recording devices. Several studies have already targeted domain generalization to improve the performance of ASC models on unseen domains, such as new devices. Recently, the Controllable Gate Adapter ConGater has been proposed in Natural Language Processing to address the biased training data problem. ConGater allows controlling the debiasing process at inference time. ConGater's main advantage is the continuous and selective debiasing of a trained model, during inference. In this work, we adapt ConGater to the audio spectrogram transformer for an acoustic scene classification task. We show that ConGater can be used to selectively adapt the learned representations to be invariant to device domain shifts such as recording devices. Our analysis shows that ConGater can progressively remove device information from the learned representations and improve the model generalization, especially under domain shift conditions (e.g. unseen devices). We show that information removal can be extended to both device and location domain. Finally, we demonstrate ConGater's ability to enhance specific device performance without further training.
Zero-shot classification of image scenes which can recognize the image scenes that are not seen in the training stage holds great promise of lowering the dependence on large numbers of labeled samples. To address the zero-shot image scene classification, the cross-modal feature alignment methods have been proposed in recent years. These methods mainly focus on matching the visual features of each image scene with their corresponding semantic descriptors in the latent space. Less attention has been paid to the contrastive relationships between different image scenes and different semantic descriptors. In light of the challenge of large intra-class difference and inter-class similarity among image scenes and the potential noisy samples, these methods are susceptible to the influence of the instances which are far from these of the same classes and close to these of other classes. In this work, we propose a multi-level cross-modal feature alignment method via contrastive learning for zero-shot classification of remote sensing image scenes. While promoting the single-instance level positive alignment between each image scene with their corresponding semantic descriptors, the proposed method takes the cross-instance contrastive relationships into consideration,and learns to keep the visual and semantic features of different classes in the latent space apart from each other. Extensive experiments have been done to evaluate the performance of the proposed method. The results show that our proposed method outperforms state of the art methods for zero-shot remote sensing image scene classification. All the code and data are available at github https://github.com/masuqiang/MCFA-Pytorch
Recent efforts have been made on acoustic scene classification in the audio signal processing community. In contrast, few studies have been conducted on acoustic scene clustering, which is a newly emerging problem. Acoustic scene clustering aims at merging the audio recordings of the same class of acoustic scene into a single cluster without using prior information and training classifiers. In this study, we propose a method for acoustic scene clustering that jointly optimizes the procedures of feature learning and clustering iteration. In the proposed method, the learned feature is a deep embedding that is extracted from a deep convolutional neural network (CNN), while the clustering algorithm is the agglomerative hierarchical clustering (AHC). We formulate a unified loss function for integrating and optimizing these two procedures. Various features and methods are compared. The experimental results demonstrate that the proposed method outperforms other unsupervised methods in terms of the normalized mutual information and the clustering accuracy. In addition, the deep embedding outperforms many state-of-the-art features.
Scene analysis is essential for enabling autonomous systems, such as mobile robots, to operate in real-world environments. However, obtaining a comprehensive understanding of the scene requires solving multiple tasks, such as panoptic segmentation, instance orientation estimation, and scene classification. Solving these tasks given limited computing and battery capabilities on mobile platforms is challenging. To address this challenge, we introduce an efficient multi-task scene analysis approach, called EMSAFormer, that uses an RGB-D Transformer-based encoder to simultaneously perform the aforementioned tasks. Our approach builds upon the previously published EMSANet. However, we show that the dual CNN-based encoder of EMSANet can be replaced with a single Transformer-based encoder. To achieve this, we investigate how information from both RGB and depth data can be effectively incorporated in a single encoder. To accelerate inference on robotic hardware, we provide a custom NVIDIA TensorRT extension enabling highly optimization for our EMSAFormer approach. Through extensive experiments on the commonly used indoor datasets NYUv2, SUNRGB-D, and ScanNet, we show that our approach achieves state-of-the-art performance while still enabling inference with up to 39.1 FPS on an NVIDIA Jetson AGX Orin 32 GB.