Abstract
In recent years, neural network based image denoising approaches have revolutionized the analysis of biomedical microscopy data. Self-supervised methods, such as Noise2Void (N2V), are applicable to virtually all noisy datasets, even without dedicated training data being available. Arguably, this facilitated the fast and widespread adoption of N2V throughout the life sciences. Unfortunately, the blind-spot training underlying N2V can lead to rather visible checkerboard artifacts, thereby reducing the quality of final predictions considerably. In this work, we present two modifications to the vanilla N2V setup that both help to reduce the unwanted artifacts considerably. Firstly, we propose a modified network architecture, i.e., using BlurPool instead of MaxPool layers throughout the used U-Net, rolling back the residual U-Net to a non-residual U-Net, and eliminating the skip connections at the uppermost U-Net level. Additionally, we propose new replacement strategies to determine the pixel intensity values that fill in the elected blind-spot pixels. We validate our modifications on a range of microscopy and natural image data. Based on added synthetic noise from multiple noise types and at varying amplitudes, we show that both proposed modifications push the current state-of-the-art for fully self-supervised image denoising.
A Point in the Right Direction: Vector Prediction for Spatially-aware Self-supervised Volumetric Representation Learning
Authors: Yejia Zhang, Pengfei Gu, Nishchal Sapkota, Hao Zheng, Peixian Liang, Danny Z. Chen
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
High annotation costs and limited labels for dense 3D medical imaging tasks have recently motivated an assortment of 3D self-supervised pretraining methods that improve transfer learning performance. However, these methods commonly lack spatial awareness despite its centrality in enabling effective 3D image analysis. More specifically, position, scale, and orientation are not only informative but also automatically available when generating image crops for training. Yet, to date, no work has proposed a pretext task that distills all key spatial features. To fulfill this need, we develop a new self-supervised method, VectorPOSE, which promotes better spatial understanding with two novel pretext tasks: Vector Prediction (VP) and Boundary-Focused Reconstruction (BFR). VP focuses on global spatial concepts (i.e., properties of 3D patches) while BFR addresses weaknesses of recent reconstruction methods to learn more effective local representations. We evaluate VectorPOSE on three 3D medical image segmentation tasks, showing that it often outperforms state-of-the-art methods, especially in limited annotation settings.
VGFlow: Visibility guided Flow Network for Human Reposing
Abstract
The task of human reposing involves generating a realistic image of a person standing in an arbitrary conceivable pose. There are multiple difficulties in generating perceptually accurate images, and existing methods suffer from limitations in preserving texture, maintaining pattern coherence, respecting cloth boundaries, handling occlusions, manipulating skin generation, etc. These difficulties are further exacerbated by the fact that the possible space of pose orientation for humans is large and variable, the nature of clothing items is highly non-rigid, and the diversity in body shape differs largely among the population. To alleviate these difficulties and synthesize perceptually accurate images, we propose VGFlow. Our model uses a visibility-guided flow module to disentangle the flow into visible and invisible parts of the target for simultaneous texture preservation and style manipulation. Furthermore, to tackle distinct body shapes and avoid network artifacts, we also incorporate a self-supervised patch-wise "realness" loss to improve the output. VGFlow achieves state-of-the-art results as observed qualitatively and quantitatively on different image quality metrics (SSIM, LPIPS, FID).
Unsupervised Feature Clustering Improves Contrastive Representation Learning for Medical Image Segmentation
Authors: Yejia Zhang, Xinrong Hu, Nishchal Sapkota, Yiyu Shi, Danny Z. Chen
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Self-supervised instance discrimination is an effective contrastive pretext task to learn feature representations and address limited medical image annotations. The idea is to make features of transformed versions of the same images similar while forcing all other augmented images' representations to contrast. However, this instance-based contrastive learning leaves performance on the table by failing to maximize feature affinity between images with similar content while counter-productively pushing their representations apart. Recent improvements on this paradigm (e.g., leveraging multi-modal data, different images in longitudinal studies, spatial correspondences) either relied on additional views or made stringent assumptions about data properties, which can sacrifice generalizability and applicability. To address this challenge, we propose a new self-supervised contrastive learning method that uses unsupervised feature clustering to better select positive and negative image samples. More specifically, we produce pseudo-classes by hierarchically clustering features obtained by an auto-encoder in an unsupervised manner, and prevent destructive interference during contrastive learning by avoiding the selection of negatives from the same pseudo-class. Experiments on 2D skin dermoscopic image segmentation and 3D multi-class whole heart CT segmentation demonstrate that our method outperforms state-of-the-art self-supervised contrastive techniques on these tasks.
Cross-Domain Self-Supervised Deep Learning for Robust Alzheimer's Disease Progression Modeling
Authors: Saba Dadsetan, Mohsen Hejrati, Shandong Wu, Somaye Hashemifar
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Developing successful artificial intelligence systems in practice depends both on robust deep learning models as well as large high quality data. Acquiring and labeling data can become prohibitively expensive and time-consuming in many real-world applications such as clinical disease models. Self-supervised learning has demonstrated great potential in increasing model accuracy and robustness in small data regimes. In addition, many clinical imaging and disease modeling applications rely heavily on regression of continuous quantities. However, the applicability of self-supervised learning for these medical-imaging regression tasks has not been extensively studied. In this study, we develop a cross-domain self-supervised learning approach for disease prognostic modeling as a regression problem using 3D images as input. We demonstrate that self-supervised pre-training can improve the prediction of Alzheimer's Disease progression from brain MRI. We also show that pre-training on extended (but not labeled) brain MRI data outperforms pre-training on natural images. We further observe that the highest performance is achieved when both natural images and extended brain-MRI data are used for pre-training.
Disentangling Task Relations for Few-shot Text Classification via Self-Supervised Hierarchical Task Clustering
Authors: Juan Zha, Zheng Li, Ying Wei, Yu Zhang
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Abstract
Few-Shot Text Classification (FSTC) imitates humans to learn a new text classifier efficiently with only few examples, by leveraging prior knowledge from historical tasks. However, most prior works assume that all the tasks are sampled from a single data source, which cannot adapt to real-world scenarios where tasks are heterogeneous and lie in different distributions. As such, existing methods may suffer from their globally knowledge-shared mechanisms to handle the task heterogeneity. On the other hand, inherent task relation are not explicitly captured, making task knowledge unorganized and hard to transfer to new tasks. Thus, we explore a new FSTC setting where tasks can come from a diverse range of data sources. To address the task heterogeneity, we propose a self-supervised hierarchical task clustering (SS-HTC) method. SS-HTC not only customizes cluster-specific knowledge by dynamically organizing heterogeneous tasks into different clusters in hierarchical levels but also disentangles underlying relations between tasks to improve the interpretability. Extensive experiments on five public FSTC benchmark datasets demonstrate the effectiveness of SS-HTC.
Fair contrastive pre-training for geographic images
Authors: Miao Zhang, Rumi Chunara
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Contrastive representation learning is widely employed in visual recognition for geographic image data (remote-sensing such as satellite imagery or proximal sensing such as street-view imagery), but because of landscape heterogeneity, models can show disparate performance across spatial units. In this work, we consider fairness risks in land-cover semantic segmentation which uses pre-trained representation in contrastive self-supervised learning. We assess class distribution shifts and model prediction disparities across selected sensitive groups: urban and rural scenes for satellite image datasets and city GDP level for a street view image dataset. We propose a mutual information training objective for multi-level latent space. The objective improves feature identification by removing spurious representations of dense local features which are disparately distributed across groups. The method achieves improved fairness results and outperforms state-of-the-art methods in terms of precision-fairness trade-off. In addition, we validate that representations learnt with the proposed method include lowest sensitive information using a linear separation evaluation. This work highlights the need for specific fairness analyses in geographic images, and provides a solution that can be generalized to different self-supervised learning methods or image data. Our code is available at: https://anonymous.4open.science/r/FairDCL-1283
Semi-Supervised and Self-Supervised Collaborative Learning for Prostate 3D MR Image Segmentation
Authors: Yousuf Babiker M. Osman, Cheng Li, Weijian Huang, Nazik Elsayed, Zhenzhen Xue, Hairong Zheng, Shanshan Wang
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Volumetric magnetic resonance (MR) image segmentation plays an important role in many clinical applications. Deep learning (DL) has recently achieved state-of-the-art or even human-level performance on various image segmentation tasks. Nevertheless, manually annotating volumetric MR images for DL model training is labor-exhaustive and time-consuming. In this work, we aim to train a semi-supervised and self-supervised collaborative learning framework for prostate 3D MR image segmentation while using extremely sparse annotations, for which the ground truth annotations are provided for just the central slice of each volumetric MR image. Specifically, semi-supervised learning and self-supervised learning methods are used to generate two independent sets of pseudo labels. These pseudo labels are then fused by Boolean operation to extract a more confident pseudo label set. The images with either manual or network self-generated labels are then employed to train a segmentation model for target volume extraction. Experimental results on a publicly available prostate MR image dataset demonstrate that, while requiring significantly less annotation effort, our framework generates very encouraging segmentation results. The proposed framework is very useful in clinical applications when training data with dense annotations are difficult to obtain.
Data Augmentation with Unsupervised Speaking Style Transfer for Speech Emotion Recognition
Authors: Leyuan Qu, Wei Wang, Taihao Li, Cornelius Weber, Stefan Wermter, Fuji Ren
Abstract
Currently, the performance of Speech Emotion Recognition (SER) systems is mainly constrained by the absence of large-scale labelled corpora. Data augmentation is regarded as a promising approach, which borrows methods from Automatic Speech Recognition (ASR), for instance, perturbation on speed and pitch, or generating emotional speech utilizing generative adversarial networks. In this paper, we propose EmoAug, a novel style transfer model to augment emotion expressions, in which a semantic encoder and a paralinguistic encoder represent verbal and non-verbal information respectively. Additionally, a decoder reconstructs speech signals by conditioning on the aforementioned two information flows in an unsupervised fashion. Once training is completed, EmoAug enriches expressions of emotional speech in different prosodic attributes, such as stress, rhythm and intensity, by feeding different styles into the paralinguistic encoder. In addition, we can also generate similar numbers of samples for each class to tackle the data imbalance issue. Experimental results on the IEMOCAP dataset demonstrate that EmoAug can successfully transfer different speaking styles while retaining the speaker identity and semantic content. Furthermore, we train a SER model with data augmented by EmoAug and show that it not only surpasses the state-of-the-art supervised and self-supervised methods but also overcomes overfitting problems caused by data imbalance. Some audio samples can be found on our demo website.
Self-supervised Egomotion and Depth Learning via Bi-directional Coarse-to-Fine Scale Recovery
Abstract
Self-supervised learning of egomotion and depth has recently attracted great attentions. These learning models can provide pose and depth maps to support navigation and perception task for autonomous driving and robots, while they do not require high-precision ground-truth labels to train the networks. However, monocular vision based methods suffer from pose scale-ambiguity problem, so that can not generate physical meaningful trajectory, and thus their applications are limited in real-world. We propose a novel self-learning deep neural network framework that can learn to estimate egomotion and depths with absolute metric scale from monocular images. Coarse depth scale is recovered via comparing point cloud data against a pretrained model that ensures the consistency of photometric loss. The scale-ambiguity problem is solved by introducing a novel two-stages coarse-to-fine scale recovery strategy that jointly refines coarse poses and depths. Our model successfully produces pose and depth estimates in global scale-metric, even in low-light condition, i.e. driving at night. The evaluation on the public datasets demonstrates that our model outperforms both representative traditional and learning based VOs and VIOs, e.g. VINS-mono, ORB-SLAM, SC-Learner, and UnVIO.
Avoid Overthinking in Self-Supervised Models for Speech Recognition
Authors: Dan Berrebbi, Brian Yan, Shinji Watanabe
Subjects: Computation and Language (cs.CL); Sound (cs.SD); Audio and Speech Processing (eess.AS)
Abstract
Self-supervised learning (SSL) models reshaped our approach to speech, language and vision. However their huge size and the opaque relations between their layers and tasks result in slow inference and network overthinking, where predictions made from the last layer of large models is worse than those made from intermediate layers. Early exit (EE) strategies can solve both issues by dynamically reducing computations at inference time for certain samples. Although popular for classification tasks in vision and language, EE has seen less use for sequence-to-sequence speech recognition (ASR) tasks where outputs from early layers are often degenerate. This challenge is further compounded when speech SSL models are applied on out-of-distribution (OOD) data. This paper first shows that SSL models do overthinking in ASR. We then motivate further research in EE by computing an optimal bound for performance versus speed trade-offs. To approach this bound we propose two new strategies for ASR: (1) we adapt the recently proposed patience strategy to ASR; and (2) we design a new EE strategy specific to ASR that performs better than all strategies previously introduced.
Masked Reconstruction Contrastive Learning with Information Bottleneck Principle
Abstract
Contrastive learning (CL) has shown great power in self-supervised learning due to its ability to capture insight correlations among large-scale data. Current CL models are biased to learn only the ability to discriminate positive and negative pairs due to the discriminative task setting. However, this bias would lead to ignoring its sufficiency for other downstream tasks, which we call the discriminative information overfitting problem. In this paper, we propose to tackle the above problems from the aspect of the Information Bottleneck (IB) principle, further pushing forward the frontier of CL. Specifically, we present a new perspective that CL is an instantiation of the IB principle, including information compression and expression. We theoretically analyze the optimal information situation and demonstrate that minimum sufficient augmentation and information-generalized representation are the optimal requirements for achieving maximum compression and generalizability to downstream tasks. Therefore, we propose the Masked Reconstruction Contrastive Learning~(MRCL) model to improve CL models. For implementation in practice, MRCL utilizes the masking operation for stronger augmentation, further eliminating redundant and noisy information. In order to alleviate the discriminative information overfitting problem effectively, we employ the reconstruction task to regularize the discriminative task. We conduct comprehensive experiments and show the superiority of the proposed model on multiple tasks, including image classification, semantic segmentation and objective detection.
Region Proposal Network Pre-Training Helps Label-Efficient Object Detection
Authors: Linus Ericsson, Nanqing Dong, Yongxin Yang, Ales Leonardis, Steven McDonagh
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Self-supervised pre-training, based on the pretext task of instance discrimination, has fueled the recent advance in label-efficient object detection. However, existing studies focus on pre-training only a feature extractor network to learn transferable representations for downstream detection tasks. This leads to the necessity of training multiple detection-specific modules from scratch in the fine-tuning phase. We argue that the region proposal network (RPN), a common detection-specific module, can additionally be pre-trained towards reducing the localization error of multi-stage detectors. In this work, we propose a simple pretext task that provides an effective pre-training for the RPN, towards efficiently improving downstream object detection performance. We evaluate the efficacy of our approach on benchmark object detection tasks and additional downstream tasks, including instance segmentation and few-shot detection. In comparison with multi-stage detectors without RPN pre-training, our approach is able to consistently improve downstream task performance, with largest gains found in label-scarce settings.
Abstract
Lifelong domain adaptation remains a challenging task in machine learning due to the differences among the domains and the unavailability of historical data. The ultimate goal is to learn the distributional shifts while retaining the previously gained knowledge. Inspired by the Complementary Learning Systems (CLS) theory, we propose a novel framework called Lifelong Self-Supervised Domain Adaptation (LLEDA). LLEDA addresses catastrophic forgetting by replaying hidden representations rather than raw data pixels and domain-agnostic knowledge transfer using self-supervised learning. LLEDA does not access labels from the source or the target domain and only has access to a single domain at any given time. Extensive experiments demonstrate that the proposed method outperforms several other methods and results in a long-term adaptation, while being less prone to catastrophic forgetting when transferred to new domains.
MAGE: MAsked Generative Encoder to Unify Representation Learning and Image Synthesis
Authors: Tianhong Li, Huiwen Chang, Shlok Kumar Mishra, Han Zhang, Dina Katabi, Dilip Krishnan
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Generative modeling and representation learning are two key tasks in computer vision. However, these models are typically trained independently, which ignores the potential for each task to help the other, and leads to training and model maintenance overheads. In this work, we propose MAsked Generative Encoder (MAGE), the first framework to unify SOTA image generation and self-supervised representation learning. Our key insight is that using variable masking ratios in masked image modeling pre-training can allow generative training (very high masking ratio) and representation learning (lower masking ratio) under the same training framework. Inspired by previous generative models, MAGE uses semantic tokens learned by a vector-quantized GAN at inputs and outputs, combining this with masking. We can further improve the representation by adding a contrastive loss to the encoder output. We extensively evaluate the generation and representation learning capabilities of MAGE. On ImageNet-1K, a single MAGE ViT-L model obtains 9.10 FID in the task of class-unconditional image generation and 78.9% top-1 accuracy for linear probing, achieving state-of-the-art performance in both image generation and representation learning. Code is available at https://github.com/LTH14/mage.
Keyword: vision transformer
Demystify Self-Attention in Vision Transformers from a Semantic Perspective: Analysis and Application
Authors: Leijie Wu, Song Guo, Yaohong Ding, Junxiao Wang, Wenchao Xu, Richard Yida Xu, Jie Zhang
Abstract
Self-attention mechanisms, especially multi-head self-attention (MSA), have achieved great success in many fields such as computer vision and natural language processing. However, many existing vision transformer (ViT) works simply inherent transformer designs from NLP to adapt vision tasks, while ignoring the fundamental difference between how MSA works in image and language settings''. Language naturally contains highly semantic structures that are directly interpretable by humans. Its basic unit (word) is discrete without redundant information, which readily supports interpretable studies on MSA mechanisms of language transformer. In contrast, visual data exhibits a fundamentally different structure: Its basic unit (pixel) is a natural low-level representation with significant redundancies in the neighbourhood, which poses obvious challenges to the interpretability of MSA mechanism in ViT. In this paper, we introduce a typical image processing technique, i.e., scale-invariant feature transforms (SIFTs), which maps low-level representations into mid-level spaces, and annotates extensive discrete keypoints with semantically rich information. Next, we construct a weighted patch interrelation analysis based on SIFT keypoints to capture the attention patterns hidden in patches with different semantic concentrations Interestingly, we find this quantitative analysis is not only an effective complement to the interpretability of MSA mechanisms in ViT, but can also be applied to 1) spurious correlation discovery andprompting'' during model inference, 2) and guided model pre-training acceleration. Experimental results on both applications show significant advantages over baselines, demonstrating the efficacy of our method.
Differentially Private Optimizers Can Learn Adversarially Robust Models
Authors: Yuan Zhang, Zhiqi Bu
Subjects: Machine Learning (cs.LG); Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
Abstract
Machine learning models have shone in a variety of domains and attracted increasing attention from both the security and the privacy communities. One important yet worrying question is: will training models under the differential privacy (DP) constraint unfavorably impact on the adversarial robustness? While previous works have postulated that privacy comes at the cost of worse robustness, we give the first theoretical analysis to show that DP models can indeed be robust and accurate, even sometimes more robust than their naturally-trained non-private counterparts. We observe three key factors that influence the privacy-robustness-accuracy tradeoff: (1) hyperparameters for DP optimizers are critical; (2) pre-training on public data significantly mitigates the accuracy and robustness drop; (3) choice of DP optimizers makes a difference. With these factors set properly, we achieve 90\% natural accuracy, 72\% robust accuracy ($+9\%$ than the non-private model) under $l2(0.5)$ attack, and 69\% robust accuracy ($+16\%$ than the non-private model) with pre-trained SimCLRv2 model under $l\infty(4/255)$ attack on CIFAR10 with $\epsilon=2$. In fact, we show both theoretically and empirically that DP models are Pareto optimal on the accuracy-robustness tradeoff. Empirically, the robustness of DP models is consistently observed on MNIST, Fashion MNIST and CelebA datasets, with ResNet and Vision Transformer. We believe our encouraging results are a significant step towards training models that are private as well as robust.
Keyword: multimodal
Navigating Connected Memories with a Task-oriented Dialog System
Abstract
Recent years have seen an increasing trend in the volume of personal media captured by users, thanks to the advent of smartphones and smart glasses, resulting in large media collections. Despite conversation being an intuitive human-computer interface, current efforts focus mostly on single-shot natural language based media retrieval to aid users query their media and re-live their memories. This severely limits the search functionality as users can neither ask follow-up queries nor obtain information without first formulating a single-turn query. In this work, we propose dialogs for connected memories as a powerful tool to empower users to search their media collection through a multi-turn, interactive conversation. Towards this, we collect a new task-oriented dialog dataset COMET, which contains $11.5k$ user<->assistant dialogs (totaling $103k$ utterances), grounded in simulated personal memory graphs. We employ a resource-efficient, two-phase data collection pipeline that uses: (1) a novel multimodal dialog simulator that generates synthetic dialog flows grounded in memory graphs, and, (2) manual paraphrasing to obtain natural language utterances. We analyze COMET, formulate four main tasks to benchmark meaningful progress, and adopt state-of-the-art language models as strong baselines, in order to highlight the multimodal challenges captured by our dataset.
Using Auxiliary Information for Person Re-Identification -- A Tutorial Overview
Authors: Tharindu Fernando, Clinton Fookes, Sridha Sridharan, Dana Michalski
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Person re-identification (re-id) is a pivotal task within an intelligent surveillance pipeline and there exist numerous re-id frameworks that achieve satisfactory performance in challenging benchmarks. However, these systems struggle to generate acceptable results when there are significant differences between the camera views, illumination conditions, or occlusions. This result can be attributed to the deficiency that exists within many recently proposed re-id pipelines where they are predominately driven by appearance-based features and little attention is paid to other auxiliary information that could aid the re-id. In this paper, we systematically review the current State-Of-The-Art (SOTA) methods in both uni-modal and multimodal person re-id. Extending beyond a conceptual framework, we illustrate how the existing SOTA methods can be extended to support these additional auxiliary information and quantitatively evaluate the utility of such auxiliary feature information, ranging from logos printed on the objects carried by the subject or printed on the clothes worn by the subject, through to his or her behavioural trajectories. To the best of our knowledge, this is the first work that explores the fusion of multiple information to generate a more discriminant person descriptor and the principal aim of this paper is to provide a thorough theoretical analysis regarding the implementation of such a framework. In addition, using model interpretation techniques, we validate the contributions from different combinations of the auxiliary information versus the original features that the SOTA person re-id models extract. We outline the limitations of the proposed approaches and propose future research directions that could be pursued to advance the area of multi-modal person re-id.
Towards Long-Tailed 3D Detection
Authors: Neehar Peri, Achal Dave, Deva Ramanan, Shu Kong
Subjects: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
Abstract
Contemporary autonomous vehicle (AV) benchmarks have advanced techniques for training 3D detectors, particularly on large-scale lidar data. Surprisingly, although semantic class labels naturally follow a long-tailed distribution, contemporary benchmarks focus on only a few common classes (e.g., pedestrian and car) and neglect many rare classes in-the-tail (e.g., debris and stroller). However, AVs must still detect rare classes to ensure safe operation. Moreover, semantic classes are often organized within a hierarchy, e.g., tail classes such as child and construction-worker are arguably subclasses of pedestrian. However, such hierarchical relationships are often ignored, which may lead to misleading estimates of performance and missed opportunities for algorithmic innovation. We address these challenges by formally studying the problem of Long-Tailed 3D Detection (LT3D), which evaluates on all classes, including those in-the-tail. We evaluate and innovate upon popular 3D detection codebases, such as CenterPoint and PointPillars, adapting them for LT3D. We develop hierarchical losses that promote feature sharing across common-vs-rare classes, as well as improved detection metrics that award partial credit to "reasonable" mistakes respecting the hierarchy (e.g., mistaking a child for an adult). Finally, we point out that fine-grained tail class accuracy is particularly improved via multimodal fusion of RGB images with LiDAR; simply put, small fine-grained classes are challenging to identify from sparse (lidar) geometry alone, suggesting that multimodal cues are crucial to long-tailed 3D detection. Our modifications improve accuracy by 5% AP on average for all classes, and dramatically improve AP for rare classes (e.g., stroller AP improves from 3.6 to 31.6)!
Cross-Mode Knowledge Adaptation for Bike Sharing Demand Prediction using Domain-Adversarial Graph Neural Networks
Abstract
For bike sharing systems, demand prediction is crucial to ensure the timely re-balancing of available bikes according to predicted demand. Existing methods for bike sharing demand prediction are mostly based on its own historical demand variation, essentially regarding it as a closed system and neglecting the interaction between different transportation modes. This is particularly important for bike sharing because it is often used to complement travel through other modes (e.g., public transit). Despite some recent progress, no existing method is capable of leveraging spatiotemporal information from multiple modes and explicitly considers the distribution discrepancy between them, which can easily lead to negative transfer. To address these challenges, this study proposes a domain-adversarial multi-relational graph neural network (DA-MRGNN) for bike sharing demand prediction with multimodal historical data as input. A temporal adversarial adaptation network is introduced to extract shareable features from demand patterns of different modes. To capture correlations between spatial units across modes, we adapt a multi-relational graph neural network (MRGNN) considering both cross-mode similarity and difference. In addition, an explainable GNN technique is developed to understand how our proposed model makes predictions. Extensive experiments are conducted using real-world bike sharing, subway and ride-hailing data from New York City. The results demonstrate the superior performance of our proposed approach compared to existing methods and the effectiveness of different model components.
Real Estate Attribute Prediction from Multiple Visual Modalities with Missing Data
Authors: Eric Stumpe, Miroslav Despotovic, Zedong Zhang, Matthias Zeppelzauer
Abstract
The assessment and valuation of real estate requires large datasets with real estate information. Unfortunately, real estate databases are usually sparse in practice, i.e., not for each property every important attribute is available. In this paper, we study the potential of predicting high-level real estate attributes from visual data, specifically from two visual modalities, namely indoor (interior) and outdoor (facade) photos. We design three models using different multimodal fusion strategies and evaluate them for three different use cases. Thereby, a particular challenge is to handle missing modalities. We evaluate different fusion strategies, present baselines for the different prediction tasks, and find that enriching the training data with additional incomplete samples can lead to an improvement in prediction accuracy. Furthermore, the fusion of information from indoor and outdoor photos results in a performance boost of up to 5% in Macro F1-score.
Keyword: CLIP
A Generalized Framework for Video Instance Segmentation
Authors: Miran Heo, Sukjun Hwang, Jeongseok Hyun, Hanjung Kim, Seoung Wug Oh, Joon-Young Lee, Seon Joo Kim
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Recently, handling long videos of complex and occluded sequences has emerged as a new challenge in the video instance segmentation (VIS) community. However, existing methods show limitations in addressing the challenge. We argue that the biggest bottleneck in current approaches is the discrepancy between the training and the inference. To effectively bridge the gap, we propose a \textbf{Gen}eralized framework for \textbf{VIS}, namely \textbf{GenVIS}, that achieves the state-of-the-art performance on challenging benchmarks without designing complicated architectures or extra post-processing. The key contribution of GenVIS is the learning strategy. Specifically, we propose a query-based training pipeline for sequential learning, using a novel target label assignment strategy. To further fill the remaining gaps, we introduce a memory that effectively acquires information from previous states. Thanks to the new perspective, which focuses on building relationships between separate frames or clips, GenVIS can be flexibly executed in both online and semi-online manner. We evaluate our methods on popular VIS benchmarks, YouTube-VIS 2019/2021/2022 and Occluded VIS (OVIS), achieving state-of-the-art results. Notably, we greatly outperform the state-of-the-art on the long VIS benchmark (OVIS), improving 5.6 AP with ResNet-50 backbone. Code will be available at https://github.com/miranheo/GenVIS.
Weakly-supervised Fingerspelling Recognition in British Sign Language Videos
Authors: K R Prajwal, Hannah Bull, Liliane Momeni, Samuel Albanie, Gül Varol, Andrew Zisserman
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
The goal of this work is to detect and recognize sequences of letters signed using fingerspelling in British Sign Language (BSL). Previous fingerspelling recognition methods have not focused on BSL, which has a very different signing alphabet (e.g., two-handed instead of one-handed) to American Sign Language (ASL). They also use manual annotations for training. In contrast to previous methods, our method only uses weak annotations from subtitles for training. We localize potential instances of fingerspelling using a simple feature similarity method, then automatically annotate these instances by querying subtitle words and searching for corresponding mouthing cues from the signer. We propose a Transformer architecture adapted to this task, with a multiple-hypothesis CTC loss function to learn from alternative annotation possibilities. We employ a multi-stage training approach, where we make use of an initial version of our trained model to extend and enhance our training data before re-training again to achieve better performance. Through extensive evaluations, we verify our method for automatic annotation and our model architecture. Moreover, we provide a human expert annotated test set of 5K video clips for evaluating BSL fingerspelling recognition methods to support sign language research.
Robust Online Video Instance Segmentation with Track Queries
Authors: Zitong Zhan, Daniel McKee, Svetlana Lazebnik
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Recently, transformer-based methods have achieved impressive results on Video Instance Segmentation (VIS). However, most of these top-performing methods run in an offline manner by processing the entire video clip at once to predict instance mask volumes. This makes them incapable of handling the long videos that appear in challenging new video instance segmentation datasets like UVO and OVIS. We propose a fully online transformer-based video instance segmentation model that performs comparably to top offline methods on the YouTube-VIS 2019 benchmark and considerably outperforms them on UVO and OVIS. This method, called Robust Online Video Segmentation (ROVIS), augments the Mask2Former image instance segmentation model with track queries, a lightweight mechanism for carrying track information from frame to frame, originally introduced by the TrackFormer method for multi-object tracking. We show that, when combined with a strong enough image segmentation architecture, track queries can exhibit impressive accuracy while not being constrained to short videos.
Keyword: metric learning
There is no result
Keyword: image retrieval
There is no result
Keyword: self-supervised
N2V2 -- Fixing Noise2Void Checkerboard Artifacts with Modified Sampling Strategies and a Tweaked Network Architecture
A Point in the Right Direction: Vector Prediction for Spatially-aware Self-supervised Volumetric Representation Learning
VGFlow: Visibility guided Flow Network for Human Reposing
Unsupervised Feature Clustering Improves Contrastive Representation Learning for Medical Image Segmentation
Cross-Domain Self-Supervised Deep Learning for Robust Alzheimer's Disease Progression Modeling
Disentangling Task Relations for Few-shot Text Classification via Self-Supervised Hierarchical Task Clustering
Fair contrastive pre-training for geographic images
Semi-Supervised and Self-Supervised Collaborative Learning for Prostate 3D MR Image Segmentation
Data Augmentation with Unsupervised Speaking Style Transfer for Speech Emotion Recognition
Self-supervised Egomotion and Depth Learning via Bi-directional Coarse-to-Fine Scale Recovery
Avoid Overthinking in Self-Supervised Models for Speech Recognition
Masked Reconstruction Contrastive Learning with Information Bottleneck Principle
Region Proposal Network Pre-Training Helps Label-Efficient Object Detection
LLEDA -- Lifelong Self-Supervised Domain Adaptation
MAGE: MAsked Generative Encoder to Unify Representation Learning and Image Synthesis
Keyword: vision transformer
Demystify Self-Attention in Vision Transformers from a Semantic Perspective: Analysis and Application
how MSA works in image and language settings''. Language naturally contains highly semantic structures that are directly interpretable by humans. Its basic unit (word) is discrete without redundant information, which readily supports interpretable studies on MSA mechanisms of language transformer. In contrast, visual data exhibits a fundamentally different structure: Its basic unit (pixel) is a natural low-level representation with significant redundancies in the neighbourhood, which poses obvious challenges to the interpretability of MSA mechanism in ViT. In this paper, we introduce a typical image processing technique, i.e., scale-invariant feature transforms (SIFTs), which maps low-level representations into mid-level spaces, and annotates extensive discrete keypoints with semantically rich information. Next, we construct a weighted patch interrelation analysis based on SIFT keypoints to capture the attention patterns hidden in patches with different semantic concentrations Interestingly, we find this quantitative analysis is not only an effective complement to the interpretability of MSA mechanisms in ViT, but can also be applied to 1) spurious correlation discovery andprompting'' during model inference, 2) and guided model pre-training acceleration. Experimental results on both applications show significant advantages over baselines, demonstrating the efficacy of our method.Differentially Private Optimizers Can Learn Adversarially Robust Models
Keyword: multimodal
Navigating Connected Memories with a Task-oriented Dialog System
Using Auxiliary Information for Person Re-Identification -- A Tutorial Overview
Towards Long-Tailed 3D Detection
Cross-Mode Knowledge Adaptation for Bike Sharing Demand Prediction using Domain-Adversarial Graph Neural Networks
Real Estate Attribute Prediction from Multiple Visual Modalities with Missing Data
Keyword: CLIP
A Generalized Framework for Video Instance Segmentation
Weakly-supervised Fingerspelling Recognition in British Sign Language Videos
Robust Online Video Instance Segmentation with Track Queries
Keyword: DALLE
There is no result