Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks276 from public data
- 1,512
- Class-Incremental Learning: Survey and Performance Evaluation on Image Classification940
This paper provides a complete survey of existing class-incremental learning methods for image classification, and in particular, it performs an extensive experimental evaluation on thirteen class-Incremental methods.
- Class-incremental learning: survey and performance evaluation940
This paper provides a complete survey of existing class-incremental learning methods for image classification, and in particular, it performs an extensive experimental evaluation on thirteen class-Incremental methods.
- 928
- The Sixth Visual Object Tracking VOT2018 Challenge Results804
The Visual Object Tracking challenge VOT2018 is the sixth annual tracker benchmarking activity organized by the VOT initiative; results of over eighty trackers are presented; many are state-of-the-art trackers published at major computer vision conferences or in journals in the recent years.
- 762
- 596
- The Visual Object Tracking VOT2014 Challenge Results588
The Visual Object Tracking challenge 2014, VOT2014, aims at comparing short-term single-object visual trackers that do not apply pre-learned models of object appearance and introduces a new dataset with full annotation of targets by rotated bounding boxes and per-frame attribute.
- RankIQA: Learning from Rankings for No-Reference Image Quality Assessment538
This work proposes a no-reference image quality assessment (NR-IQA) approach that learns from rankings (RankIQA), and demonstrates how this approach can be made significantly more efficient than traditional Siamese Networks by forward propagating a batch of images through a single network and backpropagating gradients derived from all pairs of images in the batch.
- 513
- Coloring Local Feature Extraction512
The results show that color descriptors remain reliable under photometric and geometrical changes, and with decreasing image quality, and for all experiments a combination of color and shape outperforms a pure shape-based approach.
- 496
- Semantic Drift Compensation for Class-Incremental Learning444
This work proposes a new method to estimate the drift, called semantic drift, of features and compensate for it without the need of any exemplars, and shows that the proposed SDC when combined with existing methods to prevent forgetting consistently improves results.
- Learning the Model Update for Siamese Trackers431
This work uses a convolutional neural network, called UpdateNet, which given the initial template, the accumulated template and the template of the current frame aims to estimate the optimal template for the next frame, and demonstrates the generality of the proposed approach by applying it to two Siamese trackers.
- Fast Anisotropic Gauss Filtering402
The decomposition of the anisotropic Gaussian in a one-dimensional (1-D) Gauss filter in the x-direction followed by a 1-D filter in a nonorthogonal direction phi is derived.
- Generalized Source-free Domain Adaptation342
This paper proposes a new domain adaptation paradigm called Generalized Source-free Domain Adaptation (G-SFDA), where the learned model needs to perform well on both the target and source domains, with only access to current unlabeled target data during adaptation.
- Rotate your Networks: Better Weight Consolidation and Less Catastrophic Forgetting338
An approach to avoiding catastrophic forgetting in sequential task learning scenarios based on a network reparameterization that approximately diagonalizes the Fisher Information Matrix of the network parameters, which leads to significantly better performance on lifelong learning of sequential tasks.
- 333
- Transferring GANs: Generating Images from Limited Data324
The results show that using knowledge from pretrained networks can shorten the convergence time and can significantly improve the quality of the generated images, especially when the target data is limited and it is suggested that density may be more important than diversity.
- 324
- Leveraging Unlabeled Data for Crowd Counting by Learning to Rank311
This work proposes a novel crowd counting approach that leverages abundantly available unlabeled crowd imagery in a learning-to-rank framework and demonstrates how to efficiently learn from these unlabeling datasets by incorporating learning- to-rank in a multi-task network which simultaneously ranks images and estimates crowd density maps.
- Binary patterns encoded convolutional neural networks for texture recognition and remote sensing scene classification287
This paper shows that Local Binary Patterns (LBP) encoded CNN models, codenamed TEX-Nets, trained using mapped coded images with explicit LBP based texture information provide complementary information to the standard RGB deep models.
- Generalized Gamut Mapping using Image Derivative Structures for Color Constancy241
It is analytically shown that the proposed gamut mapping framework is able to include any linear filter output and it is shown that derivatives have the advantage over pixel values to be invariant to disturbing effects (i.e. deviations of the diagonal model) such as saturated colors and diffuse light.
- Avalanche: an End-to-End Library for Continual Learning225
The proposed Avalanche, an open-source end-to-end library for continual learning research based on PyTorch is designed to provide a shared and collaborative codebase for fast prototyping, training, and reproducible evaluation of continual learning algorithms.
- Multi-Modal Fusion for End-to-End RGB-T Tracking219
This work proposes an end-to-end tracking framework for fusing the RGB and TIR modalities in RGB-T tracking, and evaluates the effectiveness of modality fusion in each of the main components in DiMP, i.e. feature extractor, target estimation network, and classifier.
- MineGAN: Effective Knowledge Transfer From GANs to Target Domains With Few Images211
This work proposes a novel knowledge transfer method for generative models based on mining the knowledge that is most beneficial to a specific target domain, either from a single or multiple pretrained GANs, and shows that it effectively transfers knowledge to domains with few target images, outperforming existing methods.
- 206
- Memory Replay GANs: Learning to Generate New Categories without Forgetting203
This paper proposes Memory Replay GANs (MeRGANs), a conditional GAN framework that integrates a memory replay generator and studies two methods to prevent forgetting by leveraging these replays, namely joint training with replay and replay alignment.
- Generative Feature Replay For Class-Incremental Learning194
Through feature generation, the method reduces the complexity of generative replay and prevents the imbalance problem and achieves state-of-the-art results on CIFAR-100 and ImageNet, while requiring only a fraction of the storage needed for exemplar-based continual learning.
- Exploiting Unlabeled Data in CNNs by Self-Supervised Learning to Rank192
The results show that networks trained to regress to the ground truth targets for labeled data and to simultaneously learn to rank unlabeled data obtain significantly better, state-of-the-art results for both IQA and crowd counting.
- 191
- 186
- Synthetic Data Generation for End-to-End Thermal Infrared Tracking183
To the best of the knowledge, this work is the first to train end-to-end features for TIR tracking, and shows that a network trained on a large dataset of synthetic TIR data obtains better performance than one trained on the available real T IR data.
- 181
- Painting-91: a large scale database for computational painting categorization176
It is shown that state-of-the-art computer vision methods can correctly classify 50 % of unseen paintings to its painter in a large dataset and correctly attribute its artistic style in over 60 % of the cases.
- Learning Object Representations for Visual Object Class Recognition171
This talk discussed the object-class recognition method that won the classification contest of the Pascal VOC Challenge 2007, and two recognition methods sharing the same underlying image representations defined by a choice of image sampler, local descriptor and global spatial grid.
- Active Learning for Deep Detection Neural Networks158
A new image-level scoring process to rank unlabeled images for their automatic selection, which clearly outperforms classical scores is proposed and can be applied to videos and sets of still images.
- 157
- 152
- 139
- NTIRE 2018 Challenge on Spectral Reconstruction from RGB Images135
This paper reviews the first challenge on spectral image reconstruction from RGB images, i.e., the recovery of whole-scene hyperspectral (HS) information from a 3-channel RGB image.
- 129
- Modulating Shape Features by Color Attention for Object Recognition125
This paper presents a novel method for recognizing object categories when using multiple cues by separately processing the shape and color cues and combining them by modulating the shape features by category-specific color attention.
- Variable Rate Deep Image Compression With Modulated Autoencoder124
Modulated autoencoders (MAEs) are proposed, where the representations of a shared autoencoder are adapted to the specific R-D tradeoff via a modulation network, and can achieve almost the same R- D performance of independent models with significantly fewer parameters.
- Learning Metrics From Teachers: Compact Networks for Image Embedding123
This paper proposes two new loss functions that model the communication of a deep teacher network to a small student network and shows that embeddings computed using small student networks perform significantly better than those computed using standard networks of similar size.
- Adversarial Networks for Spatial Context-Aware Spectral Image Reconstruction from RGB123
This work poses hyperspectral natural image reconstruction as an image to image mapping learning problem, and applies a conditional generative adversarial framework to help capture spatial semantics for the first time Convolutional Neural Networks and Generative Adversarial Networks are used to solve this task.
- Adaptive Structure Tensors and their Applications117
This chapter wants to give an overview of the different approaches to the computation of the structure tensor, whereas the focus lies on the methods based on robust statistics and nonlinear diffusion.
- Orderless Recurrent Models for Multi-Label Classification116
This paper proposes ways to dynamically order the ground truth labels with the predicted label sequence, allowing for the faster training of more optimal LSTM models for multi-label classification and outperforms other CNN-RNN models.
- Coloring Action Recognition in Still Images113
The experiments demonstrate that incorporating color information considerably improves recognition performance, and that a descriptor based on color names outperforms pure color descriptors and that late fusion of color and shape information outperforms other approaches on action recognition.
- 113
- 113
- 108
- Color in Computer Vision103
Researchers and professionals in computer science, computer vision, color science, electrical engineering, and signal processing will learn how to implement color in computer vision applications and gain insight into future developments in this dynamic and expanding field.
- Color in Computer Vision: Fundamentals and Applications103
Researchers and professionals in computer science, computer vision, color science, electrical engineering, and signal processing will learn how to implement color in computer vision applications and gain insight into future developments in this dynamic and expanding field.
- 99
- Harmony Potentials96
A new consistency potential is proposed for image labeling problems that can encode any possible combination of labels, penalizing only unlikely combinations of classes, and an effective sampling strategy is proposed over this expanded label set that renders tractable the underlying optimization problem.
- Fusing Global and Local Scale for Semantic Image Segmentation96
A new consistency potential is proposed for image labeling problems that can encode any possible combination of labels, penalizing only unlikely combinations of classes, and an effective sampling strategy is proposed over this expanded label set that renders tractable the underlying optimization problem.
- Continual Learning: Applications and the Road Forward90
This work takes a step back, and asks:"Why should one care about continual learning in the first place?", and discusses five open problems in machine learning, and shows that memory-constrained settings dominate the field.
- Unsupervised Domain Adaptation without Source Data by Casting a BAIT.89
This paper addresses Source-free Unsupervised Domain Adaptation (SFUDA), where the model has no access to any source data during the adaptation period, and proposes a novel framework named BAIT to tackle SFUDA.
- Does Multimodality Help Human and Machine for Translation and Image Captioning?88
The systems developed by LIUM and CVC for the WMT16 Multimodal Machine Translation challenge are presented, namely phrase-based systems and attentional recurrent neural networks models trained using monomodal or multimodal data.
- 85
- 85
- StyleDiffusion: Prompt-Embedding Inversion for Text-Based Editing84
Attention regularization is proposed to preserve the object-like attention maps after reconstruction and editing, enabling us to obtain accurate style editing without invoking significant structural changes.
- 83
- 81
- Reducing Label Effort: Self-Supervised meets Active Learning80
Self-training is remarkably more efficient than active learning at reducing the labeling effort, that for a low labeling budget, active learning offers no benefit to self-training, and that the combination of active learning and self- training is fruitful when the labeling budget is high.
- 76
- Review on computer vision techniques in emergency situations74
This review provides a broad overview of the progress of computer vision covering all sorts of emergencies, and examines the types of emergencies that have been studied in computer vision, the objective that the algorithms can address, the type of hardware needed and the algorithms used.
- Get What You Want, Not What You Don't: Image Content Suppression for Text-to-Image Diffusion Models73
This paper analyzes how to manipulate the text embeddings and remove unwanted content from them and introduces two contributions, which are referred to as soft-weighted regularization andference-time text embedding optimization.
- Self-Training for Class-Incremental Semantic Segmentation73
A self-training approach that leverages unlabeled data, which is used for rehearsal of previous knowledge, is applied and it is shown that maximizing self-entropy can further improve results by smoothing the overconfident predictions.
- 72
- LIUM-CVC Submissions for WMT17 Multimodal Translation Task72
The monomodal and multimodal Neural Machine Translation systems developed by LIUM and CVC for WMT17 Shared Task on Multimodal Translation ranked first for both En-De and En-Fr language pairs according to the automatic evaluation metrics METEOR and BLEU.
- 70
- Curvature estimation in oriented patterns using curvilinear models applied to gradient vector fields67
It is shown that simple curvilinear models are applicable in the analysis of a wide variety of curved oriented patterns and measure the amount of local gradient energy along the model gradient as a function of model curvature.
- Saliency of color image derivatives: a comparison between computational models and human perception66
Results from salient object detection confirm that an early fusion of color and contrast provide accurate performance to compute visual saliency and that the proposed methods are significantly better in predicting color saliency than state-of-the-art models.
- Discriminative compact pyramids for object and scene recognition65
Comparison to clustering based on agglomerative information bottleneck (AIB) shows that the method obtains superior results at significantly lower computational costs and the optimal combination of multiple features in the context of the compact pyramid representation is investigated.
- Augmented Box Replay: Overcoming Foreground Shift for Incremental Object Detection64
A novel and efficient Augmented Box Replay (ABR) method is developed that only stores and replays foreground objects and thereby circumvents the foreground shift problem and an innovative Attentive RoI Distillation loss that uses spatial attention from region-of-interest (RoI) features to constrain current model to focus on the most important information from old model.
- Domain-Adaptive Deep Network Compression64
The Domain Adaptive Low Rank (DALR) method significantly outperforms existing low-rank compression techniques and takes into account the target domain, so it can more optimally remove the redundancy in the weights.
- Continually Learning Self-Supervised Representations with Projected Functional Regularization61
It is shown that naive functional regularization leads to lower plasticity and limits continual learning performance, and proposed Projected Functional Regularization in which a separate temporal projection network ensures that the newly learned feature space preserves information of the previous one, while at the same time allowing for the learning of new features.
- 60
- Portmanteau Vocabularies for Multi-Cue Image Representation60
A novel technique for feature combination in the bag-of-words model of image classification that uses Information theoretic vocabulary compression to find discriminative combinations of cues and the resulting vocabulary of portmanteau words is compact, has the cue binding property, and supports individual weighting of cues in the final image representation.
- Saliency for fine-grained object recognition in domains with scarce training data59
The main aim of the proposed approach is to enable the effective training of a fine-grained recognition model with limited training samples and to improve the performance on the task, thereby alleviating the need to annotate a large dataset.
- 59
- Resurrecting Old Classes with New Data for Exemplar-Free Continual Learning56
This work proposes to adversarially perturb the current samples such that their embeddings are close to the old class prototypes in the old model embedding space and estimates the drift in the embedding space from the old to the new model using the perturbed images and compensates the prototypes accordingly.
- 55
- Compact color–texture description for texture classification55
This paper shows that combining multiple texture description methods significantly improves the recognition performance compared to using a single best texture method alone, and proposes to use information theoretic compression approach to compress high-dimensional multi-texture features into a compact heterogeneous texture representation.
- One-Prompt-One-Story: Free-Lunch Consistent Text-to-Image Generation Using a Single Prompt54
This paper proposes a novel training-free method for consistent text-to-image (T2I) generation, termed "One-Prompt-One-Story", which concatenates all prompts into a single input for T2I diffusion models, initially preserving character identities.
- Semi-Supervised Learning for Few-Shot Image-to-Image Translation54
This work proposes applying semi-supervised learning via a noise-tolerant pseudo-labeling procedure, and applies a cycle consistency constraint to further exploit the information from unlabeled images, either from the same dataset or external.
- 54
- 54
- Least Squares and Robust Estimation of Local Image Structure54
The robust estimators of local image structure are introduced where the quadratic error norm is replaced with a robust (Gaussian) error norm leads to a robust Gaussian facet model and examples of using them for luminance and color images are shown.
- Temporal Coherence for Active Learning in Videos53
This paper introduces a novel active learning approach for object detection in videos by exploiting temporal coherence by minimizing an energy function defined on a graphical model that provides estimates of both false positives and false negatives.
- Elastic Feature Consolidation for Cold Start Exemplar-Free Incremental Learning52
This paper considers the challenging Cold Start scenario in which insufficient data is available in the first task to learn a high-quality backbone and proposes a simple and effective approach that consolidates feature representations by regularizing drift in directions highly relevant to previous tasks and employs prototypes to reduce task-recency bias.
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference52
This paper conducts a comprehensive study of the UNet encoder and empirically analyze the encoder features, which provides insights regarding their changes during the inference process, and finds that encoder features change minimally, whereas the decoder features exhibit substantial variations across different time-steps.
- 52
- Generative Multi-Label Zero-Shot Learning51
This work is the first to tackle the problem of multi-label feature synthesis in the (generalized) zero-shot setting with a cross-level fusion-based generative approach, which outperforms the state-of-the-art on three zero- shot benchmarks: NUS-WIDE, Open Images and MS COCO.
- 49
- 46
- 46
- A Comprehensive Empirical Evaluation on Online Continual Learning45
This empirical evaluation, which focuses on the class-incremental setting in the context of image classification, finds that most methods suffer from stability and underfitting issues, however, the learned representations are comparable to i.i.d. training under the same computational budget.
- 45
- 43
- Towards Exemplar-Free Continual Learning in Vision Transformers: an Account of Attention, Functional and Weight Regularization42
This work takes an initial step towards a surgical investigation of SAM for designing coherent continual learning methods in ViTs, and acknowledges low forgetting measures for all the compared methods, indicating that ViTs might be naturally inclined continual learners.
- Class-Balanced Active Learning for Image Classification41
A general optimization framework that explicitly takes class-balancing into account is proposed that can be combined with most existing active learning algorithms and effectively applied to boost the performance of both informative and representative-based active learning methods.
- LIUM-CVC Submissions for WMT18 Multimodal Translation Task41
This paper describes the multimodal Neural Machine Translation systems developed by LIUM and CVC for WMT18 Shared Task on Multimodal Translation and proposes several modifications to the previous multimodals attention architecture in order to better integrate convolutional features and refine them using encoder-side information.
- 41
- 40
- Curvature estimation from orientation fields40
A new curvature estimator is proposed, which operates on the output of an orientation estimator, and performs well over a wide range of signal to noise ratios, down to about 0 dB.
- 38
- ICICLE: Interpretable Class Incremental Continual Learning38
This work proposes Interpretable Class-InCremental LEarning (ICICLE), an exemplar-free approach that adopts a prototypical part-based approach that reduces the interpretability concept drift and outperforms the existing exemplars-free methods of common class-incremental learning when applied to concept-based models.
- 38
- 38
- 38
- 38
- SDIT37
This work proposes SDIT: Scalable and Diverse image-to-image translation, a single generator that exploits an attention mechanism that permits the generator to focus on the domain-specific attribute.
- 36
- 35
- Mix and Match Networks: Encoder-Decoder Alignment for Zero-Pair Image Translation35
This work proposes mix and match networks, based on multiple encoders and decoders aligned in such a way that other encoder-decoder pairs can be composed at test time to perform unseen image translation tasks between domains or modalities for which explicit paired samples were not seen during training.
- Attribution-aware Weight Transfer: A Warm-Start Initialization for Class-Incremental Semantic Segmentation34
This work proposes to address the background shift with a novel classifier initialization method which employs gradient-based attribution to identify the most relevant weights for new classes from the classifier’s weights for the previous background and transfers these weights to the new classifier.
- 33
- Ternary Feature Masks: continual learning without any forgetting30
Extensive experiments on several finegrained datasets and ImageNet show that the proposed approach without any forgetting to continual learning for the task-aware regime outperforms current state-of-the-art while reducing memory overhead in comparison to weight-based approaches.
- Scale coding bag of deep features for human attribute and action recognition30
It is argued that entirely scale-invariant image representations are sub-optimal and approaches to scale coding within a bag of deep features framework are investigated and two strategies to encode multi-scale information explicitly in the final image representation are proposed.
- Combining Holistic and Part-based Deep Representations for Computational Painting Categorization29
This work presents an approach that encodes both the global scene layout and discriminative latent common structures for computational painting categorization, and achieves state-of-the-art results outperforming a recent multi-scale deep features based approach.
- Automatic Text Localisation in Scanned Comic Books29
A novel approach for the automatic text localization in scanned comics book pages, an essential step towards a fully automatic comics book understanding, focuses on speech text as it is semantically important and represents the majority of the text present in comics.
- Beyond Eleven Color Names for Image Understanding28
This work proposes a method to order the additional color names according to their complementary nature with the basic color names, which allows us to compute color name representations with high discriminative power of arbitrary length.
- Trust Your Good Friends: Source-Free Domain Adaptation by Reciprocal Neighborhood Clustering27
This paper addresses the challenging source-free domain adaptation (SFDA) problem, where the source pretrained model is adapted to the target domain in the absence of source data and demonstrates that the inherent structure of the target features is an important source of information for domain adaptation.
- Attention Distillation: self-supervised vision transformer students need more guidance27
It is shown that directly distilling information from the crucial attention mechanism from teacher to student can significantly narrow the performance gap between both self-supervised ViT andSSL methods.
- 27
- Deep spectral reflectance and illuminant estimation from self-interreflections25
This work proposes a convolutional neural network based approach to estimate the spectral reflectance of a surface and spectral power distribution of light from a single RGB image of a V-shaped surface and shows that the proposed approach outperforms state-of-the-art learning-based approaches on simulated data.
- 25
- Plasticity-Optimized Complementary Networks for Unsupervised Continual Learning24
This work proposes to train an expert network that is relieved of the duty of keeping the previous knowledge and can focus on performing optimally on the new tasks (optimizing plasticity), and shows that this approach outperforms other CURL exemplar-free methods in few- and many-task split settings.
- OneRing: A Simple Method for Source-free Open-partial Domain Adaptation24
This paper proposes a novel training scheme to learn a novel classifier to predict the n source classes and the unknown class, where samples of only known source categories are available for training and adopts a weighted entropy minimization to adapt the source pretrained model to the unlabeled target domain without source data.
- Learning to Rank for Active Learning: A Listwise Approach24
This work defines the acquisition function as a learning to rank problem and rethink the structure of the loss prediction module, using a simple but effective listwise approach, and demonstrates that this method outperforms recent state-of-the-art active learning approaches for both image classification and regression tasks.
- 24
- 24
- Continual Evidential Deep Learning for Out-of-Distribution Detection23
The integration of an evidential deep learning method into a continual learning framework in order to perform simultaneously incremental object classification and OOD detection and the ability of vacuity and dissonance to differentiate between in-distribution data belonging to old classes and OOD data is analyzed.
- Regularized Multi-Concept MIL for weakly-supervised facial behavior categorization23
This work addresses the problem of estimating high-level semantic labels for videos of recorded people by means of analysing their facial expressions and proposes a novel MIL method called Regularized Multi-Concept MIL, which outperforms state-of-the-art results in the UNBC data-set for pain detection.
- Tensor Based Feature Detection for Color Images23
This work combines feature detectors with photometric invariance theory to construct invariant features and shows that color features perform better than luminance based features and that the additional photometric information is useful to discriminate between different physical causes of features.
- One-View Occlusion Detection for Stereo Matching With a Fully Connected CRF Model22
The standard belief propagation sequential technique proposed in the tree-reweighted sequential method is extended to the fully connected conditional random field models with the geodesic distance affinity and a new approach to the BP marginal solution is proposed that is called one-view occlusion detection (OVOD).
- 22
- Improving Online Continual Learning Performance and Stability with Temporal Ensembles21
A lightweight temporal ensemble that computes the exponential moving average of the weights (EMA) at test time is used, and it is shown that it can drastically increase the performance and stability when used in combination with several methods from the literature.
- ACAE-REMIND for online continual learning with compressed feature replay21
An auxiliary classifier auto-encoder (ACAE) module is proposed for feature replay at intermediate layers with high compression rates to improve task-agnostic evaluation under online continual learning setting and get state-of-the-art performance on ImageNet-Subset, CIFar100 and CIFAR10 dataset.
- Continual learning in cross-modal retrieval21
It is found that the indexing stage pays an important role and that simply avoiding reindexing the database with updated embedding networks can lead to significant gains.
- Fusing Color and Shape for Bag-of-Words Based Object Recognition21
An analysis of existing methods for the incorporation of color in bag-of-words based image representations is provided and a list of desired properties on which bases fusing methods can be compared is proposed.
- 21
- Corner Detectors for Affine Invariant Salient Regions: Is Color Important?20
This work shows that color information can make a significant contribution to feature detection and matching and suggests that to obtain optimal performance, a tradeoff should be made between invariance and distinctiveness by an appropriate weighting of the intensity and color information.
- 19
- [Urogenital symptoms and their resulting discomfort in non-institutionalized 50-to-75-year-old Dutch women].19
The prevalence of urogenital symptoms in non-institutionalized Dutch women aged 50 to 75 years, was high: 47%.
- 18
- An Overview of Color Name Applications in Computer Vision18
An overview of color name applications in computer vision shows that in general color names outperform photometric invariants as a color representation.
- On Implicit Attribute Localization for Generalized Zero-Shot Learning17
This paper shows that common ZSL backbones can implicitly localize attributes, yet this property is not exploited, and proposes SELAR, a simple method that further encourages attribute localization, surprisingly achieving very competitive generalized ZSL (GZSL) performance when compared with more complex state-of-the-art methods.
- 17
- Image Segmentation in the Presence of Shadows and Highlights17
Qualitative results illustrate the ability of the segmentation method to obtain excellent results in the presence of shadow and highlight edges and show that it outperforms state-of-the-art segmentation methods at low computational cost.
- Calibrating Higher-Order Statistics for Few-Shot Class-Incremental Learning with Pre-trained Vision Transformers16
This work identifies the main challenge of obtaining a good covariance matrix from few-shot data and proposes to calibrate the covariance matrix for new classes based on semantic similarity to the many-shot base classes to significantly improves few-shot continual classification on several FSCIL benchmarks.
- 16
- InterLCM: Low-Quality Images as Intermediate States of Latent Consistency Models for Effective Blind Face Restoration15
Treating low-quality images as the intermediate state of LCM, InterLCM achieves a balance between fidelity and quality by starting from earlier LCM steps and incorporates a Visual Module to extract visual features and a Spatial Encoder to capture spatial details, enhancing the fidelity of restored images.
- The Art of Deception: Color Visual Illusions and Diffusion Models14
This work studies how visual illusions are encoded in diffusion models and shows that they present human-like brightness/color shifts in their latent space, demonstrating that diffusion models can predict visual illusions and showing how to generate new unseen visual illusions in realistic images using text-to-image diffusion models.
- 14
- Incremental Meta-Learning via Episodic Replay Distillation for Few-Shot Image Recognition14
This paper proposes Episodic Replay Distillation (ERD), that mixes classes from the current task with exemplars from previous tasks when sampling episodes for meta-learning to allow the training to benefit from a large as possible variety of classes, which leads to more generalizable feature representations.
- Distributed Learning and Inference With Compressed Images14
This paper sees that loss of semantic information and covariate shift do indeed exist, resulting in a drop in performance that depends on the compression rate, and proposes dataset restoration, based on image restoration with generative adversarial networks (GANs), which has the advantage of not adding additional cost to the deployed models.
- Learning Illuminant Estimation from Object Recognition14
This is the first example of a deep learning architecture for illuminant estimation that is trained without ground truth illuminants, and it is shown to outperform most deep learning methods in a cross-dataset evaluation setup, and to present competitive results in a comparison with parametric solutions.
- LocInv: Localization-aware Inversion for Text-Guided Image Editing13
Localization-aware Inversion (LocInv), which exploits segmentation maps or bounding boxes as extra localization priors to refine the cross-attention maps in the denoising phases of the diffusion process, and consistently obtains superior results both quantitatively and qualitatively.
- Articulated-Body Tracking Through Anisotropic Edge Detection13
A method that relies on both an explicit parameterization of the extremal contours and on the prediction of the human boundary edges in the image is described and a model-based anisotropic Gaussian filter is introduced to overcome the flaws of standard edge detection.
- Weakly Supervised Domain-Specific Color Naming Based on Attention12
This article adds an attention branch to the color naming network that is used to modulate the pixel-wise color naming predictions of the network and obtains state-of-the-art results forpixel-wise and image-wise classification on the EBAY dataset and is able to learn color names for various domains.
- Color Naming for Multi-color Fashion Items12
This paper focuses on color naming for images with multi-color fashion items, and shows that explicitly estimating the number of colors in the fashion item leads to improved results.
- Context proposals for saliency detection12
This paper extends object proposal methods with context proposals, which allow to incorporate the immediate context in the saliency computation, and proposes several saliency features which are computed from the context proposals.
- TEX-Nets12
This work shows that CNN models, which it is called TEX-Nets, trained using mapped coded images with explicit texture information provide complementary information to the standard deep models trained on RGB patches, and investigates two deep architectures, namely early and late fusion, to combine the texture and color information.
- 12
- Color features and local structure in images12
This dissertation aims to provide a history of web exceptionalism from 1989 to 2002, a period chosen in order to explore its roots as well as specific cases up to and including the year in which descriptions of “Web 2.0” began to circulate.
- IterInv: Iterative Inversion for Pixel-Level T2I Models11
An iterative inversion technique for large-scale text-to-image diffusion models, which employs NTI as the inversion and reconstruction of low-resolution image generation and proves the application prospects of IterInv.
- 11
- TransferI2I: Transfer Learning for Image-to-Image Translation from Small Datasets11
This paper decouple the learning process into the image generation step and the I2I translation step, and proposes two novel techniques: source-target initialization and self-initialization of the adaptor layer.
- Deep Semantic Pyramids for Human Attributes and Action Recognition11
This work states that convolutional neural networks (CNNs) or deep features have shown to improve the performance over the conventional shallow features in the context of object recognition and object detection.
- Evaluating the Impact of Color on Texture Recognition11
This paper evaluates a variety of color descriptors, popular in image classification, for texture categorisation and suggests that in all cases color names provide the best performance.
- 11
- Ask and Remember: A Questions-Only Replay Strategy for Continual Visual Question Answering10
This work presents QUestion-only replay with Attention Distillation (QUAD), a novel approach for VQACL that leverages only past task questions for regularization, and proposes Attention Consistency Distillation to enforce both intra-modal and inter-modal attention consistency across tasks, preserving essential visual-linguistic associations.
- 10
- MineGAN++: Mining Generative Models for Efficient Knowledge Transfer to Limited Data Domains10
The proposed novel knowledge transfer method, called MineGAN, effectively transfers knowledge to domains with few target images, outperforming existing methods and can successfully transfer knowledge from multiple pretrained GANs.
- Recognizing New Classes with Synthetic Data in the Loop: Application to Traffic Sign Recognition10
The results show that the proposed method to generate samples by combining synthetic images and Generative Adversarial Network technology is indeed effective, provided that a proper Convolutional Neural Network is used to perform the traffic sign recognition (classification) task as well as a proper GAN to transform the synthetic images.
- Self-supervised blur detection from synthetically blurred scenes10
This work relies on object proposals and a model for blur generation in order to produce a dataset of synthetically blurred images and performs self-supervised learning over the generated image and ground truth blur mask pairs using CNNs, defining a framework that can be employed in purely self- supervised, weakly supervised or semi-super supervised configurations.
- 9
- 9
- Mix and Match Networks: Cross-Modal Alignment for Zero-Pair Image-to-Image Translation9
This paper addresses the problem of inferring unseen cross-modal image-to-image translations between multiple modalities and proposes mix and match networks, an approach where multiple encoders and decoders are aligned in such a way that the desired translation can be obtained by simply cascading the source encoder and the target decoder, even when they have not interacted during the training stage.
- 9
- Position: Modular Memory is the Key to Continual Learning Agents8
This work outlines a conceptual framework for modular memory-centric architectures that leverage ICL for rapid adaptation and knowledge accumulation, and IWL for stable updates to model capabilities, charting a practical roadmap toward continually learning agents.
- Revisiting Weight Regularization for Low-Rank Continual Learning8
This paper revisit weight regularization in low-rank CL as a new perspective for mitigating task interference in PECL, and proposes EWC-LoRA, a practical, computational- and memory-efficient solution for CL with PTMs that achieves a stability-plasticity trade-off superior to existing low-rank CL approaches.
- 8
- 8
- Query Drift Compensation: Enabling Compatibility in Continual Learning of Retrieval Embedding Models7
This work establishes a continual learning benchmark with large-scale datasets and continually train dense retrieval embedding models on query-document pairs from new datasets in each task and observes forgetting on old tasks due to significant drift of embeddings.
- Exemplar-Free Continual Learning of Vision Transformers via Gated Class-Attention and Cascaded Feature Drift Compensation7
A new method for exemplar-free class incremental training of ViTs, which allows for plasticity towards new tasks while limiting forgetting of previous ones, and a new method of feature drift compensation that accommodates feature drift in the backbone when learning new tasks.
- 7
- 3D-Aware Multi-Class Image-to-Image Translation with NeRFs7
Quantitative and qualitative results demonstrate that the 3D-aware 121 translation with multi-view consistency is successfully performed, including a U-net-like adaptor network design, a hierarchical representation constrain and a relative regularization loss.
- Planckian Jitter: countering the color-crippling effects of color jitter on self-supervised training7
- Saliency for free: Saliency prediction as a side-effect of object recognition7
Extensive experiments carried out on both real and synthetic saliency datasets demonstrate that the approach is able to generate accurate saliency maps, achieving competitive results on both synthetic and real datasets when compared to methods that do require ground truth data.
- 7
- Exploiting the Semantic Knowledge of Pre-trained Text-Encoders for Continual Learning6
This work proposes integrating semantic guidance within and across tasks by capturing semantic similarity using text embeddings, and uses the Semantically-guided Knowledge Distillation module for enhanced knowledge transfer.
- 6
- Towards Label-Efficient Incremental Learning: A Survey6
This paper makes the first attempt to survey recently growing interest in label-efficient incremental learning, and identifies three subdivisions, namely semi-, few-shot- and self-supervised learning to reduce labeling efforts.
- 6
- Hallucinating Saliency Maps for Fine-grained Image Classification for Limited Data Domains6
This paper proposes an approach which does not require explicit saliency maps to improve image classification, but they are learned implicitely, during the training of an end-to-end image classification task, and shows that this approach obtains similar results as the case when thesaliency maps are provided explicitely.
- Controlling biases and diversity in diverse image-to-image translation6
This paper proposes the use of semantic constraints that enforce the preservation of desired image properties in image-to-image translation, and results in less unwanted changes in the translated images while still performing the wanted transformation.
- Color Edge Saliency Boosting using Natural Image Statistics6
A novel algorithm is introduced, based on the principles of independent component analysis, which models the first order derivatives of color natural images by a generalized Gaussian distribution, and it is shown that for images with a Laplacian distribution, the magnitudes of color-boosted edges reflect their corresponding information content.
- 6
- IsoCLIP: Decomposing CLIP Projectors for Efficient Intra-modal Alignment5
This paper studies intra-modal misalignment in CLIP with a focus on the role of the projectors that map pre-projection image and text embeddings into the shared embedding space and identifies an approximately isotropic subspace in which the two modalities are well-aligned.
- Transferring Unconditional to Conditional GANs with Hyper-Modulation5
This paper proposes hyper-modulated generative networks that allow for shared and complementary supervision, and introduces a self-initialization procedure that does not require any real data to initialize the hypernetwork parameters.
- HCV: Hierarchy-Consistency Verification for Incremental Implicitly-Refined Classification5
Experiments demonstrate that the proposed Hierarchy-Consistency Verification module improves performance of existing continual learning methods under this IIRC setting by a large margin.
- 5
- Object proposals for salient object segmentation in videos5
This paper investigates the application of object proposals to salient object segmentation in videos, and proposes a new motion feature derived from the optical flow structure tensor for video saliency detection.
- 4
- Covariances for Free: Exploiting Mean Distributions for Training-free Federated Learning4
This work proposes a training-free method based on an unbiased estimator of class covariance matrices which only uses first-order statistics in the form of class means communicated by clients to the server and achieves performance competitive or superior to methods sharing second-order statistics with dramatically less communication overhead.
- 4
- Density Map Distillation for Incremental Object Counting4
This paper proposes a new exemplar-free functional regularization method, called Density Map Distillation (DMD), which greatly reduces catastrophic forgetting and outperforms existing methods.
- 4
- 4
- 3D color charts for camera spectral sensitivity estimation4
This paper proposes to use spectral data of higher dimensionality by using 3D color charts that create inter-reflections between the surfaces that produce multiplications between natural spectral curves and so provide non-linear spectral curves.
- 4
- The Dichromatic Reflection Model - Future Research Directions and Applications.4
The dichromatic reflection model is resumed and two of its main applications domains are recalled: color image segmentation and photometric invariant feature computation.
- 3
- Positive Pair Distillation Considered Harmful: Continual Meta Metric Learning for Lifelong Object Re-Identification3
Based on the observation that the distillation and metric losses are antagonistic, it is proposed to remove positive pairs from distillation to robustify model updates and demonstrate that DwoPP significantly outperforms the state-of-the-art on the LReID benchmark.
- 3
- 3
- 3
- Training-free image inversion for one-step diffusion models2
Comprehensive experiments on the PIE-Bench dataset validate that the method TFinv not only achieves state-of-the-art performance in one-step diffusion editing, but also significantly outperforms existing multistep approaches in efficiency.
- 2
- AViTMP: A Tracking-Specific Transformer for Single-Branch Visual Tracking2
An Adaptive ViT Model Prediction tracker (AViTMP) is proposed to design a customised tracking method that bridges the single-branch network with discriminative models for the first time and a novel inference pipeline called CycleTrack, which bolsters the tracking robustness in the presence of distractors via bidirectional cycle tracking verification.
- Assessing Open-world Forgetting in Generative Image Model Customization2
This work presents the first systematic investigation into open-world forgetting in diffusion models, focusing on semantic and appearance drift of representations, and proposes a functional regularization strategy that effectively preserves original capabilities while accommodating new concepts.
- Visual Transformers with Primal Object Queries for Multi-Label Image Classification2
The proposed transformer model with primal object queries improves the state-of-the-art class wise F1 metric by 2.1% and 1.8%; and speeds up the convergence by 79.0% and 38.6% on MS-COCO and NUS-WIDE datasets respectively.
- 2
- Improved Discrete Optical Flow Estimation With Triple Image Matching Cost2
A new algorithm for the likelihood function calculation (alternatively the matching cost volume) that is used in the maximum a posteriori estimation of optical flow estimation is proposed, exploiting the fact that in general, optical flow is locally constant in the sense of time.
- Top-Down Deep Appearance Attention for Action Recognition2
This paper proposes a video representation which combines deep appearance and motion based local convolutional features within the bag-of-deep- features framework and outperforms both standard approaches of early and late feature fusion.
- 2
- 2
- Edge Classification for Color Constancy2
A programmable logic device (PLD) including a plurality of programmable tiles organized in blocks including a block select line is coupled to each block such that when the block select is line is asserted the address circuit of a selected block is capable of transferring data from the data bus to the plurality ofprogrammable tiles.
- 2
- LumiCtrl : Learning Illuminant Prompts for Lighting Control in Personalized Text-to-Image Models1
An illuminant personalization method named LumiCtrl that learns illuminant prompt given single image of the object and achieves significantly better illuminant fidelity, aesthetic quality, and scene coherence compared to existing baselines is presented.
- MVMO: A Multi-Object Dataset for Wide Baseline Multi-View Semantic Segmentation1
MVMO (Multi-View, Multi-Object dataset) is presented, a synthetic dataset of 116,000 scenes containing randomly placed objects of 10 distinct classes and captured from 25 camera locations in the upper hemisphere that will propel research in multi-View semantic segmentation and cross-view semantic transfer.
- Saliency from High-Level Semantic Image Features1
This work proposes several saliency features which are computed from object detection and semantic segmentation results and combines these features with a standard baseline method for saliency detection to evaluate their importance.
- 1
- Sparse Data Interpolation Using the Geodesic Distance Affinity Space1
The proposed technique is general and can be easily applied to any kind of sparse data and is compared with the popular interpolation algorithm presented in the paper on EpicFlow optical flow, which is intuitively motivated by a similar geodesic distance principle.
- Mix and match networks: multi-domain alignment for unpaired image-to-image translation1
This paper addresses the problem of inferring unseen cross-domain and cross-modal image-to-image translations between multiple domains and modalities, and proposes mix and match networks and a variant that leverages pseudo-pairs to exploit all shared information.
- Bandwidth Limited Object Recognition in High Resolution Imagery1
A novel method to optimize bandwidth usage for object detection in critical communication scenarios and develops a car recognition system via identification of its license plate and evaluates the performance of both models on a car dataset that is introduced.
- Rethinking Expert Training for Model Merging with Prompt Learning–
Dual-Tuned Experts (DTEs), a two-stage training strategy that first learns prompts and then fine-tunes the vision encoder, is introduced, a two-stage training strategy that reduces the magnitude of task-specific parameter updates and produces experts with higher merge compatibility.
- –
- –
- –
- –
- –
- Group Preference Collapse in Personalized Multimodal Large Language Models–
Experiments across multiple MLLM backbones show that PrefMoE improves preference-sensitive personalization while substantially reducing preference collapse, and routes profile and preference factors through separate LoRA adaptation paths.
- –
- Routing without Forgetting–
Routing without Forgetting is introduced, a transformer architecture augmented with energy-based associative retrieval layers inspired by Modern Hopfield Networks that indicates that embedding energy-based associative routing directly within the transformer backbone provides a principled and effective foundation for OCL.
- –
- –
- Addressing The Devastating Effects Of Single-Task Data Poisoning In Exemplar-Free Continual Learning–
It is demonstrated that even within these stringent conditions, adversaries can compromise model performance using standard image corruptions, and it is shown that STP attacks are able to strongly disrupt the whole continual training process.
- –
- –
- –
- –
- Main product detection with graph networks for fashion–
This paper proposes a model that incorporates Graph Convolutional Networks (GCN) that jointly represent all detected bounding boxes in the gallery as nodes that is better than the state-of-the-art and shows that the proposed method outperforms previous approaches by a large margin.
- –
- –
- –
- Special Section Guest Editorial:Color in Texture and Material Recognition–
This PDF file contains the editorial “Special Section Guest Editorial: color in Texture and Material Recognition” for JEI Vol.
- –
- Interactive Visual and Semantic Image Retrieval–
This chapter investigates how high-level information provided by recently developed object recognition techniques can improve interactive image retrieval and shows that automatic labeling of images with semantic labels can improve image retrieval results.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Color constant texture segmentation and retrieval–
It is concluded that color constant texture matching in image libraries provides high retrieval accuracy and is robust to varying illumination and viewing conditions.
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-14. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.