Robust Vision Research Articles

“Different New Music in Yugoslavia”: The Earliest Minimalist Manifestations

Some of Yugoslavia’s most radical musical manifestations unfolded during the 1970s and 1980s. Aided by postwar Yugoslav cultural programs, the iconic Student Cultural Center (SKC) in Belgrade – a student-managed “safe space” for free expression and creativity – opened its doors in 1971 and instantly became a magnet for experimental musicians throughout the Western and Eastern Blocs. A group of rebellious composers and performers – Opus 4 and the Ensemble for Different New Music made some of the largest strides in Yugoslav music history and defined a new era of avant-garde and a unique brand of minimalism, influenced by performance and conceptual art, Cage-style experimentalism, Fluxus, and the use of multimedia. This study briefly illustrates their robust vision that connected the East with the West in a fruitful exchange of ideas, despite the geo-political division of the world at the time.

Optical inspection of stator slots for electric motors

An optical non-contact inspection system was developed for measuring the slots in stator lamination stacks. To avoid passing go/no-go gage blocks through the slots, a machine vision system is instead used to measure the stator core slots and identify the presence of burrs within the slots. Utilizing telecentric optics along with an alignment monitoring system configured to monitor and orient the stator core, the core slots can be oriented relative to the imaging axis for further metrology measurements. Among these measurements, the smallest opening dimensions (slot width and depth) of each slot due to misalignment of laminations and the detection of burrs along the edges of the slots throughout the length of the lamination stack are critical for full stator assembly. Advanced image processing algorithms were developed to obtain sub-pixel accuracy which is required to measure the slots. This, used in conjunction with a robust vision calibration technique, increases the feasibility of building a device that can be implemented as a production inspection system. Experiments show the reliability of the computer vision approach and how it can be used in the inspection of slots in lamination stacks.

Augmented Reality Interface for Adverse-Visibility Conditions Validated by First Responders in Rescue Training Scenarios

Updating the equipment of the first responder (FR) by providing them with new capabilities and useful information will inevitably lead to better mission success rates and, therefore, more lives saved. This paper describes the design and implementation of a modular interface for augmented reality displays integrated into standard FR equipment that will provide support during the adverse-visibility situations that the rescuers find during their missions. This interface includes assistance based on the machine learning module denoted as Robust Vision Module, which detects relevant objects in a rescue scenario, particularly victims, using the feed from a thermal camera. This feed can be displayed directly alongside the detected objects, helping FRs to avoid missing anything during their operations. Additionally, the information exposition in the interface is organized according to the biometrical parameters of FRs during the operations. The main novelty of the project is its orientation towards useful solutions for FRs focusing on something occasionally ignored during research projects: the point of view of the final user. The functionalities have been designed after multiple iterations between researchers and FRs, involving testing and evaluation through realistic situations in training scenarios. Thanks to this feedback, the overall satisfaction according to the evaluations of 18 FRs is 3.84 out of 5 for the Robust Vision Module and 3.99 out of 5 for the complete AR interface. These functionalities and the different display modes available for the FRs to adapt to each situation are detailed in this paper.

DAT-Net: Deep Aggregation Transformer Network for automatic nuclear segmentation

Detection and segmentation of cell nuclei in hematoxylin and eosin-stained tissue images pose critical clinical challenges, including intricate backgrounds, variable nuclear appearances, overlapping nuclei, and unclear boundaries. These complexities render automated instance segmentation a difficult research area. Thus, this paper proposes a deep aggregation transformer network for automatic nucleus instance segmentation. The method employs a robust Vision Transformer-based backbone to extract informative features from images. To fuse multi-level information, we propose the group aggregation multi-level feature fusion module, which merges low-level features for shape representation and high-level features for pixel edge segmentation, enhancing overall model performance. The depthwise integrated multilayer perceptron module enhances feature learning by combining the global context-dependent representation of MLP and the local representation of convolution through channel separation. This fosters feature learning across different modes, thereby improving network representation. The bidirectional path interaction feature processing module is introduced for multi-scale information extraction and fusion. This module enhances the entire feature pyramid hierarchy by transferring deep semantic information to shallow layers with robust spatial information via bidirectional paths. The Large kernel Attention module at each layer extensively improves feature extraction through a large receptive field. Our approach outperforms state-of-the-art models, including StarDIST, CPP-Net, TSFD-Net and TransNuSeg, by 7.8%, 6.3%, 1.5% and 3.5% in average panoptic quality for different cell nuclei segmentation categories across tissues of the Pannuke dataset. Furthermore, DAT-Net outperforms all other state-of-the-art models in terms of the F1 score for the five classes of cell nuclei in the PanNuke dataset.

Robust vision detection of pipeline solder joints

PurposeSolder joint inspection plays a critical role in various industries, with a focus on integrated chip (IC) solder joints and metal surface welds. However, the detection of tubular solder joints has received relatively less attention. This paper aims to address the challenges of detecting small targets and complex environments by proposing a robust visual detection method for pipeline solder joints. The method is characterized by its simplicity, cost-effectiveness and ease of implementation.Design/methodology/approachA robust visual detection method based on the characteristics of pipeline solder joints is proposed. With the improved hue, saturation and value (HSV) color space, the method uses a multi-level template matching approach to first segment the pipeline from the background, and then match the endpoint of the pipeline to accurately locate the solder joint. The proposed method leverages the distinctive characteristics of pipeline solder joints and employs an enhanced HSV color space. A multi-level template matching approach is utilized to segment the pipeline from the background and accurately locate the solder joint by matching the pipeline endpoint.FindingsThe experimental results demonstrate the effectiveness of the proposed solder joint detection method in practical detection tasks. The average precision of pipeline weld joint localization exceeds 95%, while the average recall is greater than 90%. These findings highlight the applicability of the method to pipeline solder joint detection tasks, specifically in the context of production lines for refrigeration equipment.Research limitations/implicationsThe precision of the method is influenced by the placement angle and lighting conditions of the test specimen, which may pose challenges and impact the algorithm's performance. Potential avenues for improvement include exploring deep learning methods, incorporating additional features and contextual information for localization, and utilizing advanced image enhancement techniques to improve image quality.Originality/valueThe proposed pipeline solder joint detection method offers a novel and practical approach. The simplicity, cost-effectiveness and ease of implementation make it an attractive choice for detecting pipeline solder joints in different industrial applications.

Editor’s Note: Special Issue on Robust Vision

Editor’s Note: Special Issue on Robust Vision

Exploiting data diversity in multi-domain federated learning

Federated learning (FL) is an evolving machine learning technique that allows collaborative model training without sharing the original data among participants. In real-world scenarios, data residing at multiple clients are often heterogeneous in terms of different resolutions, magnifications, scanners, or imaging protocols, and thus challenging for global FL model convergence in collaborative training. Most of the existing FL methods consider data heterogeneity within one domain by assuming same data variation in each client site. In this paper, we consider data heterogeneity in FL with different domains of heterogeneous data by raising the problems of domain-shift, class-imbalance, and missing data. We propose a method, multi-domain FL as a solution to heterogeneous training data from multiple domains by training robust vision transformer model. We use two loss functions, one for correctly predicting class labels and other for encouraging similarity and dissimilarity over latent features, to optimize the global FL model. We perform various experiments using different convolution-based networks and non-convolutional Transformer architectures on multi-domain datasets. We evaluate the proposed approach on benchmark datasets and compare with the existing FL methods. Our results show the superiority of the proposed approach which performs better in term of robust FL global model than the exiting methods.

Quantitative regularization in robust vision transformer for remote sensing image classification

AbstractVision Transformers (ViTs) are exceptional at vision tasks. However, when applied to remote sensing images (RSIs), existing methods often necessitate extensive modifications of ViTs to rival convolutional neural networks (CNNs). This requirement significantly impedes the application of ViTs in geosciences, particularly for researchers who lack the time for comprehensive model redesign. To address this issue, we introduce the concept of quantitative regularization (QR), designed to enhance the performance of ViTs in RSI classification. QR represents an effective algorithm that adeptly manages domain discrepancies in RSIs and can be integrated with any ViTs in transfer learning. We evaluated the effectiveness of QR using three ViT architectures: vanilla ViT, Swin‐ViT and Next‐ViT, on four datasets: AID30, NWPU45, AFGR50 and UCM21. The results reveal that our Next‐ViT model surpasses 39 other advanced methods published in the past 3 years, maintaining robust performance even with a limited number of training samples. We also discovered that our ViT and Swin‐ViT achieve significantly higher accuracy and robustness compared to other methods using the same backbone. Our findings confirm that ViTs can be as effective as CNNs for RSI classification, regardless of the dataset size. Our approach exclusively employs open‐source ViTs and easily accessible training strategies. Consequently, we believe that our method can significantly lower the barriers for geoscience researchers intending to use ViT for RSI applications.

How Parties and Ambitious Politicians Undermined American Democracy—and the Reforms that Might Save It

Abstract Asserting that American checks and balances have broken down, Lawrence Jacobs asserts that parties must filter out dangerous candidates such as Donald Trump. To do so requires a number of reforms to undo several wrong turns in the evolution of the party system. Ambitious political elites including Jefferson, Jackson, LaFollette, and McGovern introduced changes in the name of greater equality that served their own interests but also had a number of harmful consequences. Reforms are needed to expand and reinvigorate citizen political participation along many dimensions and curb elite influence. In his book Democracy Under Fire: Donald Trump and the Breaking of American History, Lawrence Jacobs offers a robust vision of democracy and argues democracy is incompatible with entrenching minority control of electoral outcomes or political decision-making. Yet some of the failures and changes that gave rise to current threats to democracy, including authoritarian candidates, have affected other nations and have causes that go beyond remedies considered here. Some of the modest reforms proposed by Jacobs seem unlikely to achieve the purposes he envisions. However, reforms that do not require constitutional amendments and broader citizen political engagement that extends beyond election season are welcome proposals.

Mixup Virtual Adversarial Training for Robust Vision Transformers

Mixup Virtual Adversarial Training for Robust Vision Transformers

Successes and failures of conflict governance: Lessons from Aceh, Indonesia and Moro, Philippines

The purpose of this study is to analyze ethnic conflict management in Aceh (Indonesia) and Moro (Philippines) conflicts. The Aceh conflict was considered resolved after the signing of the Helsinki agreement, which was mediated by the Crisis Management Initiative (CMI), an international NGO based in Finland, on 15 August 2005 between Indonesian government and the Free Aceh Movement (Gerakan Aceh Merdeka, GAM). Meanwhile, the Moro (Philippines) conflict is still ongoing despite the TRIPOLI agreement signed and mediated by the OIC (Organization of Islamic Conference) between the Philippine government and the Moro National Liberation Front (MNLF) on 23 December 1976 and also with the MILF in 2001. This research method uses a qualitative approach. Primary data was obtained through interviews with figures who were considered competent in the Aceh conflict and the Moro conflict. The results of this study show that CMI in Aceh has succeeded in changing the situation of ethnic conflict into a situation that can reduce and reduce escalation and direct conflict actors towards conflict resolution. Peacemaking, peacekeeping and peace building have been played well by CMI involving the international community such as ASEAN and the European Union to manage the conflict. Meanwhile, in Moro the OIC failed to bring the conflict actors to solve their problems. The Philippine government lacks a robust vision, framework, and the necessary political will to effectively address the Moro conflict, which is of concern to the Organization of Islamic Cooperation (OIC).

Digging Into Uncertainty-Based Pseudo-Label for Robust Stereo Matching.

Due to the domain differences and unbalanced disparity distribution across multiple datasets, current stereo matching approaches are commonly limited to a specific dataset and generalize poorly to others. Such domain shift issue is usually addressed by substantial adaptation on costly target-domain ground-truth data, which cannot be easily obtained in practical settings. In this paper, we propose to dig into uncertainty estimation for robust stereo matching. Specifically, to balance the disparity distribution, we employ a pixel-level uncertainty estimation to adaptively adjust the next stage disparity searching space, in this way driving the network progressively prune out the space of unlikely correspondences. Then, to solve the limited ground truth data, an uncertainty-based pseudo-label is proposed to adapt the pre-trained model to the new domain, where pixel-level and area-level uncertainty estimation are proposed to filter out the high-uncertainty pixels of predicted disparity maps and generate sparse while reliable pseudo-labels to align the domain gap. Experimentally, our method shows strong cross-domain, adapt, and joint generalization and obtains 1st place on the stereo task of Robust Vision Challenge 2020. Additionally, our uncertainty-based pseudo-labels can be extended to train monocular depth estimation networks in an unsupervised way and even achieves comparable performance with the supervised methods.

What we talk about when we talk about class: discourse about the working class in the Church of England General Synod 2018–2022

ABSTRACT The Church of England struggles to talk clearly about class. This article examines discourse in the Church of England General Synod to consider both the language and themes which appear when middle-class Anglicans discuss class. It is an attempt to ‘turn the gaze’ on the powerful who allude to class whilst rarely using explicit language, preferring instead to talk about ‘the poor’, ‘disadvantaged’ and ‘deprived’. The article analyses why euphemistic reference to class are a problem, using Synod as an example. It explores three recurring themes – good news stories, leadership and church history – arguing that when these themes occur the working classes are largely presented as homogenous, lacking in agency and in need of a heroic church. The article argues that a theologically robust vision is necessary for the dismantling of classism in the Church. It outlines some promising theological starting points taken from General Synod and recent theological work by working-class theologians.

Rethinking Training Strategy in Stereo Matching.

In stereo matching, various learning-based approaches have shown impressive performance in solving traditional difficulties on multiple datasets. While most progress is obtained on a specific dataset with a dataset-specific network design, the performance on the single dataset and cross dataset affected by training strategy is often ignored. In this article, we analyze the relationship between different training strategies and performance by retraining some representative state-of-the-art methods (e.g., geometry and context network (GC-Net), pyramid stereo matching network (PSM-Net), and guided aggregation network (GA-Net), etc.). According to our research, it is surprising that the performance of networks on single or cross datasets is significantly improved by pre-training and data augmentation without any particular structure acquirement. Based on this discovery, we improve our previous non-local context attention network (NLCA-Net) to NLCA-Net v2 and train it with the novel strategy and rethink the training strategy of stereo matching concurrently. The quantitative experiments demonstrate that: 1) our model is capable of reaching top performance on both the single dataset and the multiple datasets with the same parameters in this study, which also won the 2nd place in the stereo task of the ECCV Robust vision Challenge 2020 (RVC 2020); and 2) on small datasets (e.g., KITTI, ETH3D, and Middlebury), the model's generalization and robustness are significantly affected by pre-training and data augmentation, even exceeding the network structure's influence in some cases. These observations present a challenge to the conventional wisdom of network architectures in this stage. We expect these discoveries to encourage researchers to rethink the current paradigm of ``excessive attention on the performance of a single small dataset'' in stereo matching.

Fusing vegetation index and ridge segmentation for robust vision based autonomous navigation of agricultural robots in vegetable farms

Vision based autonomous navigation is widely used for agricultural robots. However, factors such as large area of weed, discontinuous crop rows, and differences in ambient lighting condition during different plant growth stages have brought challenges to autonomous robotic navigation in farms. This paper presents a vision based method of fusing vegetation index and ridge segmentation for robust and precise extraction of navigation lines in lettuce farms. Firstly, vegetation index from the captured image is computed, and farm ridges are extracted using a semantic segmentation net. Then, vegetation index and ridge segmentation result are fused to obtain plant segmentation result. Since the method only needs to segment ridges, it does not need tedious manual labeling of vegetable plants in pixels to train plant segmentation net, yet provides accurate and reliable plant segmentation. Secondly, a modified Progressive Sample Consensus (PROSAC) algorithm and a distance filtering are proposed to fit line using the center points of plants, which effectively eliminates outliers and extract reliable and accurate center line of the current lane for autonomous navigation. Comprehensive experiments are carried out to validate the effectiveness of the method. The results show that the proposed method outperforms the conventional methods based on only vegetation index or ridge segmentation, by effectively reducing the interference caused by weeds, irregular branches and leaves, and missing rows. The proposed method runs at 10 Frames Per Second (FPS), thus satisfies real-time navigation of robots. Although the proposed method is only demonstrated in lettuce farms, it can be naturally applied to other vegetable farms, e.g. broccoli and early stage sugar beet farms.

Causal-ViT: Robust Vision Transformer by causal intervention

Artificial intelligence based on deep learning is better at improving the representation ability of models from data. However, due to the limitation of fixed receptive field, these agents are not able to provide a correct response outside the fixed receptive field. To address this problem, this paper provides a new perspective with improving the Image Recognition tasks. This study firstly constructs two extended receptive fields using structural causal model. Then, an approximate intervention method that changes the traditional likelihood prediction to predict the result of causal intervention is proposed. Finally, this study formulates the objective function to adapt the proxy training, which makes the whole model work well. Above all of these, a new Vision Transformer variant named Causal-ViT is proposed. Furthermore, rich experimental results of different tasks are reported. These results show that the proposed perspective makes a significant improvement in Image Recognition tasks. By simply plugging Causal-ViT to different sub-tasks, all of them bring the new benchmarks of themselves field, which proves our method is flexible.

A novel application of XAI in squinting models: A position paper

Artificial Intelligence, and Machine Learning especially, are becoming increasingly foundational to our collective future. Recent developments around generative models such as ChatGPT, and DALL-E represent just the tip of the iceberg in new gadgets that will change the way we live our lives. Convolutional Neural Networks (CNNs) and Transformer models are at the heart of advancements in the autonomous vehicles and health care industries as well. Yet these models, as impressive as they are, still make plenty of mistakes without justifying or explaining what aspects of the input or internal state, was responsible for the error. Often, the goal of automation is to increase throughput, processing as many tasks as possible in a short a period of time. For some use cases the cost of mistakes might be acceptable as long as production is increased above some set margin. However, in health care, autonomous vehicles, and financial applications, the cost of a mistake might have catastrophic consequences. For this reason, industries where single mistakes can be costly are less enthusiastic about early AI adoption. The field of eXplainable AI (XAI) has attracted significant attention in recent years with the goal of producing algorithms that shed light into the decision-making process of neural networks. In this paper we show how robust vision pipelines can be built using XAI algorithms with the goal of producing automated watchdogs that actively monitor the decision-making process of neural networks for signs of mistakes or ambiguous data. We call these robust vision pipelines, squinting pipelines.

MLDet: Towards efficient and accurate deep learning method for Marine Litter Detection

The relentless growth of plastic manufacturing since the 1980s has resulted in significant accumulation of litter throughout the ocean. Cleaning up this litter manually poses a major challenge, highlighting the value of autonomous underwater vehicles (AUVs) equipped with a robust vision detection algorithm. Despite prior studies, it remains challenging to employ AUVs to detect marine litter in actual maritime habitats. This paper introduces an efficient and accurate deep learning method called MLDet to cope with marine litter detection. The experimental findings show that the proposed method notably outperforms other object detectors. Furthermore, the present study comprehensively discusses the relationship between the proposed detection algorithm and recycling waste collected from marine litter. The encouraging conclusion of the study suggests that a dedicated detection algorithm is a reliable tool for automatically recognizing marine litter and sustaining a healthy ocean.

Technologies Supporting Screening Oculomotor Problems: Challenges for Virtual Reality

Oculomotor dysfunctions (OMDs) are problems relating to coordination and accuracy of eye movements for processing visual information. Eye-tracking (ET) technologies show great promise in the identification of OMDs. However, current computer technologies for vision screening are specialized devices with limited screen size and the inability to measure depth, while visual field and depth are important information for detecting OMDs. In this experimental study, we examine the possibilities of immersive virtual reality (VR) technologies compared with laptop technologies for increased user experiences, presence, immersiveness, and the use of serious games for identifying OMDs. The results present increased interest in VR-based screening, motivating users to focus better using VR applications free from outside distractions. These limitations currently include lower performance and confidence in results of identifying OMDs with the used HMDs. Using serious games for screening in VR is also estimated to have great potential for developing a more robust vision screening tool, especially for younger children.

Women as Bangles, Bangles as Women: Migration, Gender, and Female Agency in Bangladeshi Writer Nashid Kamal’s The Glass Bangles

Migration, women, and agency are central research topics in the world literature in English. These topics have fascinated literary researchers since the late twentieth century and enabled debate on how and to what extent migration from post-colonial countries to the West shapes women’s sense of selfhood and identity. The article joins that debate, engaging with Nashid Kamal’s The Glass Bangles, a lesser-known Bangladeshi novel in English. It focuses on the rural, middle-class, and disadvantaged heroine Sheila. It examines how the novel represents the cultural dynamics of diaspora and what images of gendered migration the heroine’s translocal and transnational mobility reflect. The article frames the discussion with Judith Butler’s “gender and precarity” and Elora Shehabuddin’s “subaltern rationality.” It works on the textual metaphor of the glass bangles and their implications, as metaphors are powerful rhetorical devices to reflect migrants’ struggles, pains, and hopes. Accordingly, the article contends that, although romantic and, thus, debatable, Kamal’s novel deserves appreciation for presenting a robust vision of precarious women’s resilience and empowerment.