Test-time adaptation updates pretrained models on the fly to handle distribution shifts in test data. While existing research has focused on stable optimization during adaptation, less attention has been given to enhancing model representations for adaptation capability. To address this gap, we propose Test-Time Ensemble (TTE) grounded in the intriguing property of linear mode connectivity. TTE leverages ensemble strategies during adaptation: 1) adaptively averaging the parameter weights of assorted test-time adapted models and 2) incorporating dropout to further promote representation diversity. These strategies encapsulate model diversity into a single model, avoiding computational burden associated with managing multiple models. Besides, we propose a robust knowledge distillation scheme to prevent model collapse, ensuring stable optimization and preserving the ensemble benefits during adaptation. Notably, TTE integrates seamlessly with existing TTA approaches, advancing their adaptation capabilities. In extensive experiments, integration with TTE consistently outperformed baseline models across various challenging scenarios, demonstrating its effectiveness and general applicability.

Neural networks are often prone to bias toward spurious correlations inherent in a dataset, thus failing to generalize unbiased test criteria. A key challenge to resolving the issue is the significant lack of bias-conflicting training data (i. e., samples without spurious correlations). In this paper, we propose a novel data augmentation approach termed Bias-Adversarial augmentation (BiasAdv) that supplements bias-conflicting samples with adversarial images. Our key idea is that an adversarial attack on a biased model that makes decisions based on spurious correlations may generate syn-thetic bias-conflicting samples, which can then be used as augmented training data for learning a debiased model. Specifically, we formulate an optimization problem for gen-erating adversarial images that attack the predictions of an auxiliary biased model without ruining the predictions of the desired debiased model. Despite its simplicity, we find that BiasAdv can generate surprisingly useful synthetic bias-conflicting samples, allowing the debiased model to learn generalizable representations. Furthermore, BiasAdv does not require any bias annotations or prior knowledge of the bias type, which enables its broad applicability to existing debiasing methods to improve their performances. Our extensive experimental results demonstrate the superiority of BiasAdv, achieving state-of-the-art performance on four popular benchmark datasets across various bias domains.

With the continual expansion of face datasets, feature-based distillation prevails for large-scale face recognition. In this work, we attempt to remove identity supervision in student training, to spare the GPU memory from saving massive class centers. However, this naive removal leads to inferior distillation result. We carefully inspect the performance degradation from the perspective of intrinsic dimension, and argue that the gap in intrinsic dimension, namely the intrinsic gap, is intimately connected to the infamous capacity gap problem. By constraining the teacher's search space with reverse distillation, we narrow the intrinsic gap and unleash the potential of feature-only distillation. Remarkably, the proposed reverse distillation creates universally student-friendly teacher that demonstrates outstanding student improvement. We further enhance its effectiveness by designing a student proxy to better bridge the intrinsic gap. As a result, the proposed method surpasses state-of-the-art distillation techniques with identity supervision on various face recognition benchmarks, and the improvements are consistent across different teacher-student pairs.

Deep learning algorithms require large amounts of labeled data for effective performance, but the presence of noisy labels often significantly degrade their performance. Although recent studies on designing a robust objective function to label noise, known as the robust loss method, have shown promising results for learning with noisy labels, they suffer from the issue of underfitting not only noisy samples but also clean ones, leading to suboptimal model performance. To address this issue, we propose a novel learning framework that selectively suppresses noisy samples while avoiding underfitting clean data. Our framework incorporates label confidence as a measure of label noise, enabling the network model to prioritize the training of samples deemed to be noise-free. The label confidence is based on the robust loss methods, and we provide theoretical evidence that our method can reach the optimal point of the robust loss, subject to certain conditions. Furthermore, the proposed method is generalizable and can be combined with existing robust loss methods, making it suitable for a wide range of applications of learning with noisy labels. We evaluate our approach on both synthetic and real-world datasets, and the experimental results demonstrate its effectiveness in achieving outstanding classification performance compared to state-of-the-art methods.

The ill-posed nature of monocular 3D geometry (depth map and surface normals) estimation makes it rely mostly on data-driven approaches such as Deep Neural Networks (DNN). However, data acquisition of surface normals, especially the reliable normals, is acknowledged difficult. Commonly, reconstruction of surface normals with high quality is heuristic and time-consuming. Such fact urges methodologies to minimize dependency on ground-truth normals when predicting 3D geometry. In this work, we devise CO-planarity REgularized (CORE) loss functions and Structure-Aware Normal Estimator (SANE). Without involving any knowledge of ground-truth normals, these two designs enable pixel-wise 3D geometry estimation weakly supervised by only ground-truth depth map. For CORE loss functions, the key idea is to exploit locally linear depth-normal orthogonality under spherical coordinates as pixel-level constraints, and utilize our designed Adaptive Polar Regularization (APR) to resolve underlying numerical degeneracies. Meanwhile, SANE easily establishes multi-task learning with CORE loss functions on both depth and surface normal estimation, leading to the whole performance leap. Extensive experiments present the effectiveness of our method on various DNN architectures and data benchmarks. The experimental results demonstrate that our depth estimation achieves the state-of-the-art performance across all metrics on indoor scenes and comparable performance on outdoor scenes. In addition, our surface normal estimation is overall superior.

In real-world applications, a single instance could have more than one label. To solve this task, multi-label learning methods emerged in recent years. It is a more challenging problem for many reasons, such as complex label correlation, long-tail label distribution, and data shortage. In general, overcoming these challenges and bettering learning performance could be achieved by utilizing more training samples and including label correlations. However, these solutions are expensive and inflexible. Large-scale, well-labeled datasets are difficult to obtain, and building label correlation maps requires task-specific semantic information as prior knowledge. To address these limitations, we propose a general and compact Multi-Label Correlation Learning (MUCO) framework. MUCO explicitly and effectively learns the latent label correlations by updating a label correlation tensor, which provides highly accurate and interpretable prediction results. In addition, a multi-label generative strategy is deployed to handle the long-tail label distribution challenge. It borrows the visual clues from limited samples and synthesizes more diverse samples. All networks in our model are optimized simultaneously. Extensive experiments illustrate the effectiveness and efficiency of MUCO. Ablation studies further prove the effectiveness of all the modules.

Accurate facial landmarks are essential prerequisites for many tasks related to human faces. In this paper, an accurate facial landmark detector is proposed based on cascaded transformers. We formulate facial landmark detection as a coordinate regression task such that the model can be trained end-to-end. With self-attention in transformers, our model can inherently exploit the structured relationships between landmarks, which would benefit landmark detection under challenging conditions such as large pose and occlusion. During cascaded refinement, our model is able to extract the most relevant image features around the target landmark for coordinate prediction, based on deformable attention mechanism, thus bringing more accurate alignment. In addition, we propose a novel decoder that refines image features and landmark positions simultaneously. With few parameter increasing, the detection performance improves further. Our model achieves new state-of-the-art performance on several standard facial landmark detection benchmarks, and shows good generalization ability in cross-dataset evaluation.

Scene text recognition (STR) attracts much attention over the years because of its wide application. Most methods train STR model in a fully supervised manner which requires large amounts of labeled data. Although synthetic data contributes a lot to STR, it suffers from the real-to-synthetic domain gap the restricts model performance. In this work, we aim to boost STR models by leveraging both synthetic data and the numerous real unlabeled images, exempting human annotation cost thoroughly. A robust con-sistency regularization based semi-supervised framework is proposed for STR, which can effectively solve the instability issue due to domain inconsistency between synthetic and real images. A character-level consistency regularization is designed to mitigate the misalignment between characters in sequence recognition. Extensive experiments on standard text recognition benchmarks demonstrate the effectiveness of the proposed method. It can steadily improve existing STR models, and boost an STR model to achieve new state-of-the-art results. To our best knowledge, this is the first consistency regularization based framework that applies successfully to STR.

Despite the remarkable performance of deep models on image recognition tasks, they are known to be susceptible to common corruptions such as blur, noise, and low-resolution. Data augmentation is a conventional way to build a robust model by considering these common corruptions during the training. However, a naive data augmentation scheme may result in a non-specialized model for particular corruptions, as the model tends to learn the averaged distribution among corruptions. To mitigate the issue, we propose a new paradigm of training deep image recognition networks that produce clean-like features from any quality image via an invertible neural architecture. The proposed method consists of two stages. In the first stage, we train an invertible network with only clean images under the recognition objective. In the second stage, its inversion, i.e., the invertible decoder, is attached to a new recognition network and we train this encoder-decoder network using both clean and corrupted images by considering recognition and reconstruction objectives. Our two-stage scheme allows the network to produce clean-like and robust features from any quality images, by reconstructing their clean images via the invertible decoder. We demonstrate the effectiveness of our method on image classification and face recognition tasks.

Humans have the ability to robustly recognize objects with various factors of variations such as nonrigid transformations, background noises, and changes in lighting conditions. However, training deep learning models generally require huge amount of data instances under diverse variations, to ensure its robustness. To alleviate the need of collecting large amount of data and better learn to generalize with scarce data instances, we propose a novel meta-learning method which learns to transfer factors of variations from one class to another, such that it can improve the classification performance on unseen examples. Transferred variations generate virtual samples that augment the feature space of the target class during training, simulating upcoming query samples with similar variations. By sharing the factors of variations across different classes, the model becomes more robust to variations in the unseen examples and tasks using small number of examples per class. We validate our model on multiple benchmark datasets for few-shot classification and face recognition, on which our model significantly improves the performance of the base model, outperforming relevant baselines.

The goal of Multi-label learning is to predict multiple labels of each single instance. This is a challenging problem since the training data is limited, long-tail label distribution, and complicated label correlations. Generally, more training samples and label correlation knowledge would benefit the learning performance. However, it is difficult to obtain large-scale well-labeled datasets, and building such a label correlation map requires sophisticated semantic knowledge. To this end, we propose an end-to-end Generative Correlation Discovery Network (GCDN) method for multi-label learning in this paper. GCDN captures the existing data distribution, and synthesizes diverse data to enlarge the diversity of the training features; meanwhile, it also learns the label correlations based on a specifically-designed, simple but effective correlation discovery network to automatically discover the label correlations and considerately improve the label prediction accuracy. Extensive experiments on several benchmarks are provided to demonstrate the effectiveness, efficiency, and high accuracy of our approach.

While interacting with mobile devices, users may press against touch screens and also exert tangential force to the display in a sliding manner. We seek to guide UI design based on the tangential force applied by a user to the surface of a hand-held device. A prototype of an interface using tangential force input was implemented utilizing a force sensitive layer and an elastic layer and used for the user experiment. We investigated user controllability to reach and maintain target force levels and considered the effects of hand pose and direction of force input. Our results imply no significant difference in performance when applying force holding the device in one hand and in two hands. We also observed that users have more physical and perceived loads when applying tangential force in the left-right direction compared to the up-down direction. Based on the experimental results, we discuss considerations for user interface applications of tangential-force-based interface.

The recent online networked virtual worlds such as SecondLife, World of Warcraft and Lineage have been increasingly popular. A life-scale virtual world presentation and the intuitive interaction between the users and the virtual worlds would provide more natural and immersive experience for users. The emergence of novel interaction technologies such as sensing the facial expression and the motion of the users and the real world environments could be used to provide a strong connection between them. For the wide acceptance and use of the virtual world, a various type of novel interaction devices should have a unified interaction formats between the real world and the virtual world and interoperability among virtual worlds. Thus, MPEG-V Media Context and Control (ISO/IEC 23005) standardizes such connecting information. The paper provides an overview and its usage example of MPEG-V from the real world to the virtual world (R2V) on interfaces for controlling avatars and virtual objects in the virtual world by the real world devices. In particular, we investigate how the MPEG-V framework can be applied for the facial animation of an avatar in various types of virtual worlds.

In this paper, we implement the newly introduced correntropy MACE filter using the fast Gauss transform (FGT). The correntropy MACE filter is a nonlinear extension to the MACE filter using the correntropy function in a feature space nonlinearly related to the input. The correntropy MACE outperforms the traditional linear MACE in both generalization and rejection abilities. However, in practice, the drawback of the correntropy MACE filter is its computation complexity. This paper present a fast version of the correntropy MACE by using the FGT idea and validates the approximation with results in synthetic aperture radar (SAR) image recognition.

Switching approaches can improve the performance of adaptive schemes, however a data driven criterion to accomplish the task is unclear. In this paper, we propose a new optimization criterion for switching which is estimated directly from data. We apply the method to the recently introduced MEE and MEE-SAS algorithms. Using this novel switching scheme, we develop a single algorithm which effectively combines the strengths of MEE and MEE-SAS without sacrificing the simplicity and stability properties of MEE. We explain these results analytically, and through simulations.

In this paper, we revisit information theoretic vector quantization (ITVQ) algorithm introduced in (T. Lehn-Schioler et al., 2005) and make it practical. We derive a fixed point update rule to minimize the Cauchy-Schwartz(CS) pdf divergence between the set of codewords and the actual data. In doing so, we overcome two severe deficiencies of the previous gradient based method namely, the number of parameters to be optimized and slow convergence rate, thus making this algorithm more efficient and useful as a compression algorithm.

We propose in this paper the normalized Minimum Error Entropy (NMEE). Following the same rational that lead to the normalized LMS, the weight update adjustment for Minimum Error Entropy (MEE) is constrained by the principle of minimum disturbance. Unexpectedly, we obtained an algorithm that not only is insensitive to the power of the input, but is also faster than the MEE for the same misadjustment, and also that is less sensitive to the kernel size. We explain these results analytically, and through system identification simulations.

In this paper a non-linear extension to the synthetic discriminant function (SDF) is proposed. The SDF is a well known 2-D correlation filter for object recognition. The proposed non-linear version of the SDF is derived from kernel-based learning. The kernel SDF is implemented in a nonlinear high dimensional space by using the kernel trick and it can improve the performance of the linear SDF by incorporating the image's class higher order moments. We show that this kernelized composite correlation filter has an intrinsic connection with the recently proposed correntropy function. We apply this kernel SDF to face recognition and simulations show that the kernel SDF significantly outperforms the traditional SDF as well as is robust in noisy data environments.

We propose a new method of clustering neural spike waveforms for spike sorting. After detecting the spikes using a threshold detector, we use principal component analysis (PCA) to get the first few PCA components of the data. Clustering on these PCA components is achieved by maximizing the Cauchy Schwartz PDF divergence measure which uses the Parzen window method to non parametrically estimate the pdf of the clusters. Comparison with other clustering techniques in spike sorting like k-means and Gaussian mixture elucidates the superiority of our method in terms of classification results and computational complexity.