We consider class incremental learning (CIL) problem, in which a learning agent continuously learns new classes from incrementally arriving training data batches and aims to predict well on all the classes learned so far. The main challenge of the problem is the catastrophic forgetting, and for the exemplar-memory based CIL methods, it is generally known that the forgetting is commonly caused by the classification score bias that is injected due to the data imbalance between the new classes and the old classes (in the exemplar-memory). While several methods have been proposed to correct such score bias by some additional post-processing, e.g., score re-scaling or balanced fine-tuning, no systematic analysis on the root cause of such bias has been done. To that end, we analyze that computing the softmax probabilities by combining the output scores for all old and new classes could be the main cause of the bias. Then, we propose a new CIL method, dubbed as Separated Softmax for Incremental Learning (SS-IL), that consists of separated softmax (SS) output layer combined with task-wise knowledge distillation (TKD) to resolve such bias. Throughout our extensive experimental results on several large-scale CIL benchmark datasets, we show our SS-IL achieves strong state-of-the-art accuracy through attaining much more balanced prediction scores across old and new classes, without any additional post-processing.
论文检索
输入标题、作者或关键词，从 7,876 篇学术成果中精准定位
Recent studies have shown that many deep metric learning loss functions perform very similarly under the same experimental conditions. One potential reason for this unexpected result is that all losses let the network focus on similar image regions or properties. In this paper, we investigate this by conducting a two-step analysis to extract and compare the learned visual features of the same model architecture trained with different loss functions: First, we compare the learned features on the pixel level by correlating saliency maps of the same input images. Second, we compare the clustering of embeddings for several image properties, e.g. object color or illumination. To provide independent control over these properties, photo-realistic 3D car renders similar to images in the Cars196 dataset are generated. In our analysis, we compare 14 pretrained models from a recent study and find that, even though all models perform similarly, different loss functions can guide the model to learn different features. We especially find differences between classification and ranking based losses. Our analysis also shows that some seemingly irrelevant properties can have significant influence on the resulting embedding. We encourage researchers from the deep metric learning community to use our methods to get insights into the features learned by their proposed methods.
We investigate the impact of aliasing on generalization in Deep Convolutional Networks and show that data augmentation schemes alone are unable to prevent it due to structural limitations in widely used architectures. Drawing insights from frequency analysis theory, we take a closer look at Resnet and EfficientNet architectures and review the trade-off between aliasing and information loss in each of their major components. We show how to mitigate aliasing by inserting non-trainable low-pass filters at key locations, particularly where networks lack the capacity to learn them. These simple architectural changes lead to substantial improvements in generalization on i.i.d. and even more on out-of-distribution conditions, such as image classification under natural corruptions on ImageNet-C and few-shot learning on Meta-Dataset. State-of-the art results are achieved on both datasets without introducing additional trainable parameters and using the default hyper-parameters of open source codebases.
In this paper, we present Co-scale conv-attentional image Transformers (CoaT), a Transformer-based image classifier equipped with co-scale and conv-attentional mechanisms. First, the co-scale mechanism maintains the integrity of Transformers' encoder branches at individual scales, while allowing representations learned at different scales to effectively communicate with each other; we design a series of serial and parallel blocks to realize the co-scale mechanism. Second, we devise a conv-attentional mechanism by realizing a relative position embedding formulation in the factorized attention module with an efficient convolution-like implementation. CoaT empowers image Transformers with enriched multi-scale and contextual modeling capabilities. On ImageNet, relatively small CoaT models attain superior classification results compared with similar-sized convolutional neural networks and image/vision Transformers. The effectiveness of CoaT's backbone is also illustrated on object detection and instance segmentation, demonstrating its applicability to downstream computer vision tasks.
Compact convolutional neural networks (CNNs) have witnessed exceptional improvements in performance in recent years. However, they still fail to provide the same predictive power as CNNs with a large number of parameters. The diverse and even abundant features captured by the layers is an important characteristic of these successful CNNs. However, differences in this characteristic between large CNNs and their compact counterparts have rarely been investigated. In compact CNNs, due to the limited number of parameters, abundant features are unlikely to be obtained, and feature diversity becomes an essential characteristic. Diverse features present in the activation maps derived from a data point during model inference may indicate the presence of a set of unique descriptors necessary to distinguish between objects of different classes. In contrast, data points with low feature diversity may not provide a sufficient amount of unique descriptors to make a valid prediction; we refer to them as random predictions. Random predictions can negatively impact the optimization process and harm the final performance. This paper proposes addressing the problem raised by random predictions by reshaping the standard cross-entropy to make it biased toward data points with a limited number of unique descriptive features. Our novel Bias Loss focuses the training on a set of valuable data points and prevents the vast number of samples with poor learning features from misleading the optimization process. Furthermore, to show the importance of diversity, we present a family of SkipblockNet models whose architectures are brought to boost the number of unique descriptors in the last layers. Experiments conducted on benchmark datasets demonstrate the superiority of the proposed loss function over the cross-entropy loss. Moreover, our SkipblockNet-M can achieve 1% higher classification accuracy than MobileNetV3 Large with similar computational cost on the ImageNet ILSVRC-2012 classification dataset. The code is available on the link - https://github.com/lusinlu/biasloss_skipblocknet.
For several emerging technologies such as augmented reality, autonomous driving and robotics, visual localization is a critical component. Directly regressing camera pose/3D scene coordinates from the input image using deep neural networks has shown great potential. However, such methods assume a stationary data distribution with all scenes simultaneously available during training. In this paper, we approach the problem of visual localization in a continual learning setup -- whereby the model is trained on scenes in an incremental manner. Our results show that similar to the classification domain, non-stationary data induces catastrophic forgetting in deep networks for visual localization. To address this issue, a strong baseline based on storing and replaying images from a fixed buffer is proposed. Furthermore, we propose a new sampling method based on coverage score (Buff-CS) that adapts the existing sampling strategies in the buffering process to the problem of visual localization. Results demonstrate consistent improvements over standard buffering methods on two challenging datasets -- 7Scenes, 12Scenes, and also 19Scenes by combining the former scenes.
Transformer architectures have become the model of choice in natural language processing and are now being introduced into computer vision tasks such as image classification, object detection, and semantic segmentation. However, in the field of human pose estimation, convolutional architectures still remain dominant. In this work, we present PoseFormer, a purely transformer-based approach for 3D human pose estimation in videos without convolutional architectures involved. Inspired by recent developments in vision transformers, we design a spatial-temporal transformer structure to comprehensively model the human joint relations within each frame as well as the temporal correlations across frames, then output an accurate 3D human pose of the center frame. We quantitatively and qualitatively evaluate our method on two popular and standard benchmark datasets: Human3.6M and MPI-INF-3DHP. Extensive experiments show that PoseFormer achieves state-of-the-art performance on both datasets. Our code and model will be publicly available.
Recent breakthroughs in self-supervised learning show that such algorithms learn visual representations that can be transferred better to unseen tasks than cross-entropy based methods which rely on task-specific supervision. In this paper, we found that the similar holds in the continual learning context: contrastively learned representations are more robust against the catastrophic forgetting than ones trained with the cross-entropy objective. Based on this novel observation, we propose a rehearsal-based continual learning algorithm that focuses on continually learning and maintaining transferable representations. More specifically, the proposed scheme (1) learns representations using the contrastive learning objective, and (2) preserves learned representations using a self-supervised distillation step. We conduct extensive experimental validations under popular benchmark image classification datasets, where our method sets the new state-of-the-art performance. Source code is available at https://github.com/chaht01/Co2L.
Egocentric video recognition is a challenging task that requires to identify both the actor's motion and the active object that the actor interacts with. Recognizing the active object is particularly hard due to the cluttered background with distracting objects, the frequent field of view changes, severe occlusion, etc. To improve the active object classification, most existing methods use object detectors or human gaze information, which are computationally expensive or require labor-intensive annotations. To avoid these additional costs, we propose an end-to-end Interactive Prototype Learning (IPL) framework to learn better active object representations by leveraging the motion cues from the actor. First, we introduce a set of verb prototypes to disentangle active object features from distracting object features. Each prototype corresponds to a primary motion pattern of an egocentric action, offering a distinctive supervision signal for active object feature learning. Second, we design two interactive operations to enable the extraction of active object features, i.e., noun-to-verb assignment and verb-to-noun selection. These operations are parameter-efficient and can learn judicious location-aware features on top of 3D CNN backbones. We demonstrate that the IPL framework can generalize to different backbones and outperform the state-of-the-art on three large-scale egocentric video datasets, i.e., EPIC-KITCHENS-55, EPIC-KITCHENS-100 and EGTEA.
Compared with traditional methods, the deep learning-based multi-focus image fusion methods can effectively improve the performance of image fusion tasks. However, the existing deep learning-based methods encounter a common issue of a large number of parameters, which leads to the deep learning models with high time complexity and low fusion efficiency. To address this issue, we propose a novel discrete Tchebichef moment-based Deep neural network, termed as DTMNet, for multi-focus image fusion. The proposed DTMNet is an end-to-end deep neural network with only one convolutional layer and three fully connected layers. The convolutional layer is fixed with DTM coefficients (DTMConv) to extract high/low-frequency information without learning parameters effectively. The three fully connected layers have learnable parameters for feature classification. Therefore, the proposed DTMNet for multi-focus image fusion has a small number of parameters (0.01M paras vs. 4.93M paras of regular CNN) and high computational efficiency (0.32s vs. 79.09s by regular CNN to fuse an image). In addition, a large-scale multi-focus image dataset is synthesized for training and verifying the deep learning model. Experimental results on three public datasets demonstrate that the proposed method is competitive with or even outperforms the state-of-the-art multi-focus image fusion methods in terms of subjective visual perception and objective evaluation metrics.
Recent methods for long-tailed instance segmentation still struggle on rare object classes with few training data. We propose a simple yet effective method, Feature Augmentation and Sampling Adaptation (FASA), that addresses the data scarcity issue by augmenting the feature space especially for rare classes. Both the Feature Augmentation (FA) and feature sampling components are adaptive to the actual training status -- FA is informed by the feature mean and variance of observed real samples from past iterations, and we sample the generated virtual features in a loss-adapted manner to avoid over-fitting. FASA does not require any elaborate loss design, and removes the need for inter-class transfer learning that often involves large cost and manually-defined head/tail class groups. We show FASA is a fast, generic method that can be easily plugged into standard or long-tailed segmentation frameworks, with consistent performance gains and little added cost. FASA is also applicable to other tasks like long-tailed classification with state-of-the-art performance.
Event cameras are bio-inspired sensors that respond to brightness changes asynchronously and output in the form of event streams instead of frame-based images. They own outstanding advantages compared with traditional cameras: higher temporal resolution, higher dynamic range, and lower power consumption. However, the spatial resolution of existing event cameras is insufficient and challenging to be enhanced at the hardware level while maintaining the asynchronous philosophy of circuit design. Therefore, it is imperative to explore the algorithm of event stream super-resolution, which is a non-trivial task due to the sparsity and strong spatio-temporal correlation of the events from an event camera. In this paper, we propose an end-to-end framework based on spiking neural network for event stream super-resolution, which can generate high-resolution (HR) event stream from the input low-resolution (LR) event stream. A spatiotemporal constraint learning mechanism is proposed to learn the spatial and temporal distributions of the event stream simultaneously. We validate our method on four large-scale datasets and the results show that our method achieves state-of-the-art performance. The satisfying results on two downstream applications, i.e. object classification and image reconstruction, further demonstrate the usability of our method. To prove the application potential of our method, we deploy it on a mobile platform. The high-quality HR event stream generated by our real-time system demonstrates the effectiveness and efficiency of our method.
Generalizing deep neural networks to new target domains is critical to their real-world utility. In practice, it may be feasible to get some target data labeled, but to be cost-effective it is desirable to select a maximally-informative subset via active learning (AL). We study the problem of AL under a domain shift, called Active Domain Adaptation (Active DA). We demonstrate how existing AL approaches based solely on model uncertainty or diversity sampling are less effective for Active DA. We propose Clustering Uncertainty-weighted Embeddings (CLUE), a novel label acquisition strategy for Active DA that performs uncertainty-weighted clustering to identify target instances for labeling that are both uncertain under the model and diverse in feature space. CLUE consistently outperforms competing label acquisition strategies for Active DA and AL across learning settings on 6 diverse domain shifts for image classification.
Face clustering plays an essential role in exploiting massive unlabeled face data. Recently, graph-based face clustering methods are getting popular for their satisfying performances. However, they usually suffer from excessive memory consumption especially on large-scale graphs, and rely on empirical thresholds to determine the connectivities between samples in inference, which restricts their applications in various real-world scenes. To address such problems, in this paper, we explore face clustering from the pairwise angle. Specifically, we formulate the face clustering task as a pairwise relationship classification task, avoiding the memory-consuming learning on large-scale graphs. The classifier can directly determine the relationship between samples and is enhanced by taking advantage of the contextual information. Moreover, to further facilitate the efficiency of our method, we propose a rank-weighted density to guide the selection of pairs sent to the classifier. Experimental results demonstrate that our method achieves state-of-the-art performances on several public clustering benchmarks at the fastest speed and shows a great advantage in comparison with graph-based clustering methods on memory consumption.
HIRE-SNN: Harnessing the Inherent Robustness of Energy-Efficient Deep Spiking Neural Networks by Training With Crafted Input Noise
PDF ↗Low-latency deep spiking neural networks (SNNs) have become a promising alternative to conventional artificial neural networks (ANNs) because of their potential for increased energy efficiency on event-driven neuromorphic hardware. Neural networks, including SNNs, however, are subject to various adversarial attacks and must be trained to remain resilient against such attacks for many applications. Nevertheless, due to prohibitively high training costs associated with SNNs, analysis, and optimization of deep SNNs under various adversarial attacks have been largely overlooked. In this paper, we first present a detailed analysis of the inherent robustness of low-latency SNNs against popular gradient-based attacks, namely fast gradient sign method (FGSM) and projected gradient descent (PGD). Motivated by this analysis, to harness the model robustness against these attacks we present an SNN training algorithm that uses crafted input noise and incurs no additional training time. To evaluate the merits of our algorithm, we conducted extensive experiments with variants of VGG and ResNet on both CIFAR-10 and CIFAR-100 datasets. Compared to standard trained direct input SNNs, our trained models yield improved classification accuracy of up to 13.7% and 10.1% on FGSM and PGD attack-generated images, respectively, with negligible loss in clean image accuracy. Our models also outperform inherently-robust SNNs trained on rate-coded inputs with improved or similar classification performance on attack-generated im-ages while having up to 25x and 4.6x lower latency and computation energy, respectively.
Generalized and Incremental Few-Shot Learning by Explicit Learning and Calibration Without Forgetting
PDF ↗Both generalized and incremental few-shot learning have to deal with three major challenges: learning novel classes from only few samples per class, preventing catastrophic forgetting of base classes, and classifier calibration across novel and base classes. In this work we propose a three-stage framework that allows to explicitly and effectively address these challenges. While the first phase learns base classes with many samples, the second phase learns a calibrated classifier for novel classes from few samples while also preventing catastrophic forgetting. In the final phase, calibration is achieved across all classes. We evaluate the proposed framework on four challenging benchmark datasets for image and video few-shot classification and obtain state-of-the-art results for both generalized and incremental few shot learning.
Deep neural networks have achieved remarkable performance on a range of classification tasks, with softmax cross-entropy (CE) loss emerging as the de-facto objective function. The CE loss encourages features of a class to have a higher projection score on the true class-vector compared to the negative classes. However, this is a relative constraint and does not explicitly force different class features to be well-separated. Motivated by the observation that ground-truth class representations in CE loss are orthogonal (one-hot encoded vectors), we develop a novel loss function termed `Orthogonal Projection Loss' (OPL) which imposes orthogonality in the feature space. OPL augments the properties of CE loss and directly enforces inter-class separation alongside intra-class clustering in the feature space through orthogonality constraints on the mini-batch level. As compared to other alternatives of CE, OPL offers unique advantages e.g., no additional learnable parameters, does not require careful negative mining and is not sensitive to the batch size. Given the plug-and-play nature of OPL, we evaluate it on a diverse range of tasks including image recognition (CIFAR-100), large-scale classification (ImageNet), domain generalization (PACS) and few-shot learning (miniImageNet, CIFAR-FS, tiered-ImageNet and Meta-dataset) and demonstrate its effectiveness across the board. Furthermore, OPL offers better robustness against practical nuisances such as adversarial attacks and label noise. Our code will be publicly released.
Few-Shot Image Classification: Just Use a Library of Pre-Trained Feature Extractors and a Simple Classifier
PDF ↗Recent papers have suggested that transfer learning can outperform sophisticated meta-learning methods for few-shot image classification. We take this hypothesis to its logical conclusion, and suggest the use of an ensemble of high-quality, pre-trained feature extractors for few-shot image classification. We show experimentally that a library of pre-trained feature extractors combined with a simple feed-forward network learned with an L2-regularizer can be an excellent option for solving cross-domain few-shot image classification. Our experimental results suggest that this simpler sample-efficient approach far outperforms several well-established meta-learning algorithms.
We consider the deep point cloud analysis tasks where the inputs of the networks are randomly rotated. Recent progress in rotation-invariant point cloud analysis is mainly driven by converting point clouds into their respective canonical poses, and principal component analysis (PCA) is a practical tool to achieve this. Due to the imperfect alignment of PCA, most of the current works are devoted to developing powerful network structures and features to overcome this deficiency, without thoroughly analyzing the PCA-based canonical poses themselves. In this work, we present a detailed study w.r.t. the PCA-based canonical poses of point clouds. Our investigation reveals that the ambiguity problem associated with the PCA-based canonical poses is handled insufficiently in some recent works. To this end, we develop a simple pose selector module for disambiguation, which presents noticeable enhancement (i.e., 5:3% classification accuracy) over state-of-the-art approaches on the challenging real-world dataset.
Vis2Mesh: Efficient Mesh Reconstruction From Unstructured Point Clouds of Large Scenes With Learned Virtual View Visibility
PDF ↗We present a novel framework for mesh reconstruction from unstructured point clouds by taking advantage of the learned visibility of the 3D points in the virtual views and traditional graph-cut based mesh generation. Specifically, we first propose a three-step network that explicitly employs depth completion for visibility prediction. Then the visibility information of multiple views is aggregated to generate a 3D mesh model by solving an optimization problem considering visibility in which a novel adaptive visibility weighting term in surface determination is also introduced to suppress line of sight with a large incident angle. Compared to other learning-based approaches, our pipeline only exercises the learning on a 2D binary classification task, i.e., points visible or not in a view, which is much more generalizable and practically more efficient and capable to deal with a large number of points. Experiments demonstrate that our method with favorable transferability and robustness, and achieve competing performances w.r.t. state-of-the-art learning-based approaches on small complex objects and outperforms on large indoor and outdoor scenes. Code is available at https://github.com/GDAOSU/vis2mesh.