PyTorchKR Papers

최신 논문

총 10,163편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2109.12909공식 코드 · Papers with Code

Compressive Visual Representations

Learning effective visual representations that generalize well without human supervision is a fundamental problem in order to apply Machine Learning to a wide variety of tasks. Recently, two families of self-supervised methods, contrastive learning and latent bootstrapping, exemplified by SimCLR and BYOL respectively, have made significant progress. In this work, we hypothesize that adding explicit information compression to these algorithms yields better and more robust representations. We verify this by developing SimCLR and BYOL formulations compatible with the Conditional Entropy Bottleneck (CEB) objective, allowing us to both measure and control the amount of compression in the learned representation, and observe their impact on downstream tasks. Furthermore, we explore the relationship between Lipschitz continuity and compression, showing a tractable lower bound on the Lipschitz constant of the encoders we learn. As Lipschitz continuity is closely related to robustness, this provides a new explanation for why compressed models are more robust. Our experiments confirm that adding compression to SimCLR and BYOL significantly improves linear evaluation accuracies and model robustness across a wide range of domain shifts. In particular, the compressed version of BYOL achieves 76.0% Top-1 linear evaluation accuracy on ImageNet with ResNet-50, and 78.8% with ResNet-50 2x.

arXiv:2109.12556

Frequency Disentangled Residual Network

Residual networks (ResNets) have been utilized for various computer vision and image processing applications. The residual connection improves the training of the network with better gradient flow. A residual block consists of few convolutional layers having trainable parameters, which leads to overfitting. Moreover, the present residual networks are not able to utilize the high and low frequency information suitably, which also challenges the generalization capability of the network. In this paper, a frequency disentangled residual network (FDResNet) is proposed to tackle these issues. Specifically, FDResNet includes separate connections in the residual block for low and high frequency components, respectively. Basically, the proposed model disentangles the low and high frequency components to increase the generalization ability. Moreover, the computation of low and high frequency components using fixed filters further avoids the overfitting. The proposed model is tested on benchmark CIFAR10/100, Caltech and TinyImageNet datasets for image classification. The performance of the proposed model is also tested in image retrieval framework. It is noticed that the proposed model outperforms its counterpart residual model. The effect of kernel size and standard deviation is also evaluated. The impact of the frequency disentangling is also analyzed using saliency map.

arXiv:2109.12548공식 코드 · Papers with Code

Disentangled Feature Representation for Few-shot Image Classification

Learning the generalizable feature representation is critical for few-shot image classification. While recent works exploited task-specific feature embedding using meta-tasks for few-shot learning, they are limited in many challenging tasks as being distracted by the excursive features such as the background, domain and style of the image samples. In this work, we propose a novel Disentangled Feature Representation framework, dubbed DFR, for few-shot learning applications. DFR can adaptively decouple the discriminative features that are modeled by the classification branch, from the class-irrelevant component of the variation branch. In general, most of the popular deep few-shot learning methods can be plugged in as the classification branch, thus DFR can boost their performance on various few-shot tasks. Furthermore, we propose a novel FS-DomainNet dataset based on DomainNet, for benchmarking the few-shot domain generalization tasks. We conducted extensive experiments to evaluate the proposed DFR on general and fine-grained few-shot classification, as well as few-shot domain generalization, using the corresponding four benchmarks, i.e., mini-ImageNet, tiered-ImageNet, CUB, as well as the proposed FS-DomainNet. Thanks to the effective feature disentangling, the DFR-based few-shot classifiers achieved the state-of-the-art results on all datasets.

arXiv:2109.12453

Classification of COVID-19 from CXR Images in a 15-class Scenario: an Attempt to Avoid Bias in the System

As of June 2021, the World Health Organization (WHO) has reported 171.7 million confirmed cases including 3,698,621 deaths from COVID-19. Detecting COVID-19 and other lung diseases from Chest X-Ray (CXR) images can be very effective for emergency diagnosis and treatment as CXR is fast and cheap. The objective of this study is to develop a system capable of detecting COVID-19 along with 14 other lung diseases from CXRs in a fair and unbiased manner. The proposed system consists of a CXR image selection technique and a deep learning based model to classify 15 diseases including COVID-19. The proposed CXR selection technique aims to retain the maximum variation uniformly and eliminate poor quality CXRs with the goal of reducing the training dataset size without compromising classifier accuracy. More importantly, it reduces the often hidden bias and unfairness in decision making. The proposed solution exhibits a promising COVID-19 detection scheme in a more realistic situation than most existing studies as it deals with 15 lung diseases together. We hope the proposed method will have wider adoption in medical image classification and other related fields.

arXiv:2109.12338

Distribution-sensitive Information Retention for Accurate Binary Neural Network

Model binarization is an effective method of compressing neural networks and accelerating their inference process. However, a significant performance gap still exists between the 1-bit model and the 32-bit one. The empirical study shows that binarization causes a great loss of information in the forward and backward propagation. We present a novel Distribution-sensitive Information Retention Network (DIR-Net) that retains the information in the forward and backward propagation by improving internal propagation and introducing external representations. The DIR-Net mainly relies on three technical contributions: (1) Information Maximized Binarization (IMB): minimizing the information loss and the binarization error of weights/activations simultaneously by weight balance and standardization; (2) Distribution-sensitive Two-stage Estimator (DTE): retaining the information of gradients by distribution-sensitive soft approximation by jointly considering the updating capability and accurate gradient; (3) Representation-align Binarization-aware Distillation (RBD): retaining the representation information by distilling the representations between full-precision and binarized networks. The DIR-Net investigates both forward and backward processes of BNNs from the unified information perspective, thereby providing new insight into the mechanism of network binarization. The three techniques in our DIR-Net are versatile and effective and can be applied in various structures to improve BNNs. Comprehensive experiments on the image classification and objective detection tasks show that our DIR-Net consistently outperforms the state-of-the-art binarization approaches under mainstream and compact architectures, such as ResNet, VGG, EfficientNet, DARTS, and MobileNet. Additionally, we conduct our DIR-Net on real-world resource-limited devices which achieves 11.1x storage saving and 5.4x speedup.

arXiv:2109.12271공식 코드 · Papers with Code

BiTr-Unet: a CNN-Transformer Combined Network for MRI Brain Tumor Segmentation

Convolutional neural networks (CNNs) have achieved remarkable success in automatically segmenting organs or lesions on 3D medical images. Recently, vision transformer networks have exhibited exceptional performance in 2D image classification tasks. Compared with CNNs, transformer networks have an appealing advantage of extracting long-range features due to their self-attention algorithm. Therefore, we propose a CNN-Transformer combined model, called BiTr-Unet, with specific modifications for brain tumor segmentation on multi-modal MRI scans. Our BiTr-Unet achieves good performance on the BraTS2021 validation dataset with median Dice score 0.9335, 0.9304 and 0.8899, and median Hausdorff distance 2.8284, 2.2361 and 1.4142 for the whole tumor, tumor core, and enhancing tumor, respectively. On the BraTS2021 testing dataset, the corresponding results are 0.9257, 0.9350 and 0.8874 for Dice score, and 3, 2.2361 and 1.4142 for Hausdorff distance. The code is publicly available at https://github.com/JustaTinyDot/BiTr-Unet.

arXiv:2109.12040

From images in the wild to video-informed image classification

Image classifiers work effectively when applied on structured images, yet they often fail when applied on images with very high visual complexity. This paper describes experiments applying state-of-the-art object classifiers toward a unique set of images in the wild with high visual complexity collected on the island of Bali. The text describes differences between actual images in the wild and images from Imagenet, and then discusses a novel approach combining informational cues particular to video with an ensemble of imperfect classifiers in order to improve classification results on video sourced images of plants in the wild.

arXiv:2109.11839

Frequency Pooling: Shift-Equivalent and Anti-Aliasing Downsampling

Convolution utilizes a shift-equivalent prior of images, thus leading to great success in image processing tasks. However, commonly used poolings in convolutional neural networks (CNNs), such as max-pooling, average-pooling, and strided-convolution, are not shift-equivalent. Thus, the shift-equivalence of CNNs is destroyed when convolutions and poolings are stacked. Moreover, anti-aliasing is another essential property of poolings from the perspective of signal processing. However, recent poolings are neither shift-equivalent nor anti-aliasing. To address this issue, we propose a new pooling method that is shift-equivalent and anti-aliasing, named frequency pooling. Frequency pooling first transforms the features into the frequency domain, and then removes the frequency components beyond the Nyquist frequency. Finally, it transforms the features back to the spatial domain. We prove that frequency pooling is shift-equivalent and anti-aliasing based on the property of Fourier transform and Nyquist frequency. Experiments on image classification show that frequency pooling improves accuracy and robustness with respect to the shifts of CNNs.

arXiv:2109.11806

A Multi-stage Transfer Learning Framework for Diabetic Retinopathy Grading on Small Data

Diabetic retinopathy (DR) is one of the major blindness-causing diseases currently known. Automatic grading of DR using deep learning methods not only speeds up the diagnosis of the disease but also reduces the rate of misdiagnosis. However,problems such as insufficient samples and imbalanced class distribution in small DR datasets have constrained the improvement of grading performance. In this paper, we apply the idea of multi-stage transfer learning into the grading task of DR. The new transfer learning technique utilizes multiple datasets with different scales to enable the model to learn more feature representation information. Meanwhile, to cope with the imbalanced problem of small DR datasets, we present a class-balanced loss function in our work and adopt a simple and easy-to-implement training method for it. The experimental results on IDRiD dataset show that our method can effectively improve the grading performance on small data, obtaining scores of 0.7961 and 0.8763 in terms of accuracy and quadratic weighted kappa, respectively. Our method also outperforms several state-of-the-art methods.

arXiv:2109.11249공식 코드 · Papers with Code

FooBaR: Fault Fooling Backdoor Attack on Neural Network Training

Neural network implementations are known to be vulnerable to physical attack vectors such as fault injection attacks. As of now, these attacks were only utilized during the inference phase with the intention to cause a misclassification. In this work, we explore a novel attack paradigm by injecting faults during the training phase of a neural network in a way that the resulting network can be attacked during deployment without the necessity of further faulting. In particular, we discuss attacks against ReLU activation functions that make it possible to generate a family of malicious inputs, which are called fooling inputs, to be used at inference time to induce controlled misclassifications. Such malicious inputs are obtained by mathematically solving a system of linear equations that would cause a particular behaviour on the attacked activation functions, similar to the one induced in training through faulting. We call such attacks fooling backdoors as the fault attacks at the training phase inject backdoors into the network that allow an attacker to produce fooling inputs. We evaluate our approach against multi-layer perceptron networks and convolutional networks on a popular image classification task obtaining high attack success rates (from 60% to 100%) and high classification confidence when as little as 25 neurons are attacked while preserving high accuracy on the originally intended classification task.

arXiv:2109.10582공식 코드 · Papers with Code

Partial sensitivity analysis in differential privacy

Differential privacy (DP) allows the quantification of privacy loss when the data of individuals is subjected to algorithmic processing such as machine learning, as well as the provision of objective privacy guarantees. However, while techniques such as individual R\'enyi DP (RDP) allow for granular, per-person privacy accounting, few works have investigated the impact of each input feature on the individual's privacy loss. Here we extend the view of individual RDP by introducing a new concept we call partial sensitivity, which leverages symbolic automatic differentiation to determine the influence of each input feature on the gradient norm of a function. We experimentally evaluate our approach on queries over private databases, where we obtain a feature-level contribution of private attributes to the DP guarantee of individuals. Furthermore, we explore our findings in the context of neural network training on synthetic data by investigating the partial sensitivity of input pixels on an image classification task.

arXiv:2109.09952

Multi-Domain Few-Shot Learning and Dataset for Agricultural Applications

Automatic classification of pests and plants (both healthy and diseased) is of paramount importance in agriculture to improve yield. Conventional deep learning models based on convolutional neural networks require thousands of labeled examples per category. In this work we propose a method to learn from a few samples to automatically classify different pests, plants, and their diseases, using Few-Shot Learning (FSL). We learn a feature extractor to generate embeddings and then update the embeddings using Transformers. Using Mahalanobis distance, a class-covariance-based metric, we then calculate the similarity of the transformed embeddings with the embedding of the image to be classified. Using our proposed architecture, we conduct extensive experiments on multiple datasets showing the effectiveness of our proposed model. We conduct 42 experiments in total to comprehensively analyze the model and it achieves up to 14% and 24% performance gains on few-shot image classification benchmarks on two datasets. We also compile a new FSL dataset containing images of healthy and diseased plants taken in real-world settings. Using our proposed architecture which has been shown to outperform several existing FSL architectures in agriculture, we provide strong baselines on our newly proposed dataset.

arXiv:2109.09850공식 코드 · Papers with Code

Balanced-MixUp for Highly Imbalanced Medical Image Classification

Highly imbalanced datasets are ubiquitous in medical image classification problems. In such problems, it is often the case that rare classes associated to less prevalent diseases are severely under-represented in labeled databases, typically resulting in poor performance of machine learning algorithms due to overfitting in the learning process. In this paper, we propose a novel mechanism for sampling training data based on the popular MixUp regularization technique, which we refer to as Balanced-MixUp. In short, Balanced-MixUp simultaneously performs regular (i.e., instance-based) and balanced (i.e., class-based) sampling of the training data. The resulting two sets of samples are then mixed-up to create a more balanced training distribution from which a neural network can effectively learn without incurring in heavily under-fitting the minority classes. We experiment with a highly imbalanced dataset of retinal images (55K samples, 5 classes) and a long-tail dataset of gastro-intestinal video frames (10K images, 23 classes), using two CNNs of varying representation capabilities. Experimental results demonstrate that applying Balanced-MixUp outperforms other conventional sampling schemes and loss functions specifically designed to deal with imbalanced data. Code is released at https://github.com/agaldran/balanced_mixup .

arXiv:2109.09495공식 코드 · Papers with Code

GhostShiftAddNet: More Features from Energy-Efficient Operations

Deep convolutional neural networks (CNNs) are computationally and memory intensive. In CNNs, intensive multiplication can have resource implications that may challenge the ability for effective deployment of inference on resource-constrained edge devices. This paper proposes GhostShiftAddNet, where the motivation is to implement a hardware-efficient deep network: a multiplication-free CNN with fewer redundant features. We introduce a new bottleneck block, GhostSA, that converts all multiplications in the block to cheap operations. The bottleneck uses an appropriate number of bit-shift filters to process intrinsic feature maps, then applies a series of transformations that consist of bit-wise shifts with addition operations to generate more feature maps that fully learn to capture information underlying intrinsic features. We schedule the number of bit-shift and addition operations for different hardware platforms. We conduct extensive experiments and ablation studies with desktop and embedded (Jetson Nano) devices for implementation and measurements. We demonstrate the proposed GhostSA block can replace bottleneck blocks in the backbone of state-of-the-art networks architectures and gives improved performance on image classification benchmarks. Further, our GhostShiftAddNet can achieve higher classification accuracy with fewer FLOPs and parameters (reduced by up to 3x) than GhostNet. When compared to GhostNet, inference latency on the Jetson Nano is improved by 1.3x and 2x on the GPU and CPU respectively. Code is available open-source on \url{https://github.com/JIABI/GhostShiftAddNet}.

arXiv:2109.08937공식 코드 · Papers with Code

UNetFormer: A UNet-like Transformer for Efficient Semantic Segmentation of Remote Sensing Urban Scene Imagery

Semantic segmentation of remotely sensed urban scene images is required in a wide range of practical applications, such as land cover mapping, urban change detection, environmental protection, and economic assessment.Driven by rapid developments in deep learning technologies, the convolutional neural network (CNN) has dominated semantic segmentation for many years. CNN adopts hierarchical feature representation, demonstrating strong capabilities for local information extraction. However, the local property of the convolution layer limits the network from capturing the global context. Recently, as a hot topic in the domain of computer vision, Transformer has demonstrated its great potential in global information modelling, boosting many vision-related tasks such as image classification, object detection, and particularly semantic segmentation. In this paper, we propose a Transformer-based decoder and construct a UNet-like Transformer (UNetFormer) for real-time urban scene segmentation. For efficient segmentation, the UNetFormer selects the lightweight ResNet18 as the encoder and develops an efficient global-local attention mechanism to model both global and local information in the decoder. Extensive experiments reveal that our method not only runs faster but also produces higher accuracy compared with state-of-the-art lightweight models. Specifically, the proposed UNetFormer achieved 67.8% and 52.4% mIoU on the UAVid and LoveDA datasets, respectively, while the inference speed can achieve up to 322.4 FPS with a 512x512 input on a single NVIDIA GTX 3090 GPU. In further exploration, the proposed Transformer-based decoder combined with a Swin Transformer encoder also achieves the state-of-the-art result (91.3% F1 and 84.1% mIoU) on the Vaihingen dataset. The source code will be freely available at https://github.com/WangLibo1995/GeoSeg.

arXiv:2109.09536

Audio-Visual Speech Recognition is Worth 32$\times$32$\times$8 Voxels

Audio-visual automatic speech recognition (AV-ASR) introduces the video modality into the speech recognition process, often by relying on information conveyed by the motion of the speaker's mouth. The use of the video signal requires extracting visual features, which are then combined with the acoustic features to build an AV-ASR system [1]. This is traditionally done with some form of 3D convolutional network (e.g. VGG) as widely used in the computer vision community. Recently, image transformers [2] have been introduced to extract visual features useful for image classification tasks. In this work, we propose to replace the 3D convolutional visual front-end with a video transformer front-end. We train our systems on a large-scale dataset composed of YouTube videos and evaluate performance on the publicly available LRS3-TED set, as well as on a large set of YouTube videos. On a lip-reading task, the transformer-based front-end shows superior performance compared to a strong convolutional baseline. On an AV-ASR task, the transformer front-end performs as well as (or better than) the convolutional baseline. Fine-tuning our model on the LRS3-TED training set matches previous state of the art. Thus, we experimentally show the viability of the convolution-free model for AV-ASR.

arXiv:2109.09389

Explaining Convolutional Neural Networks by Tagging Filters

Convolutional neural networks (CNNs) have achieved astonishing performance on various image classification tasks, but it is difficult for humans to understand how a classification comes about. Recent literature proposes methods to explain the classification process to humans. These focus mostly on visualizing feature maps and filter weights, which are not very intuitive for non-experts in analyzing a CNN classification. In this paper, we propose FilTag, an approach to effectively explain CNNs even to non-experts. The idea is that when images of a class frequently activate a convolutional filter, then that filter is tagged with that class. These tags provide an explanation to a reference of a class-specific feature detected by the filter. Based on the tagging, individual image classifications can then be intuitively explained in terms of the tags of the filters that the input image activates. Finally, we show that the tags are helpful in analyzing classification errors caused by noisy input images and that the tags can be further processed by machines.

arXiv:2109.09063

Ontology-based n-ball Concept Embeddings Informing Few-shot Image Classification

We propose a novel framework named ViOCE that integrates ontology-based background knowledge in the form of $n$-ball concept embeddings into a neural network based vision architecture. The approach consists of two components - converting symbolic knowledge of an ontology into continuous space by learning n-ball embeddings that capture properties of subsumption and disjointness, and guiding the training and inference of a vision model using the learnt embeddings. We evaluate ViOCE using the task of few-shot image classification, where it demonstrates superior performance on two standard benchmarks.

arXiv:2109.09016

The Unreasonable Effectiveness of the Final Batch Normalization Layer

Early-stage disease indications are rarely recorded in real-world domains, such as Agriculture and Healthcare, and yet, their accurate identification is critical in that point of time. In this type of highly imbalanced classification problems, which encompass complex features, deep learning (DL) is much needed because of its strong detection capabilities. At the same time, DL is observed in practice to favor majority over minority classes and consequently suffer from inaccurate detection of the targeted early-stage indications. In this work, we extend the study done by Kocaman et al., 2020, showing that the final BN layer, when placed before the softmax output layer, has a considerable impact in highly imbalanced image classification problems as well as undermines the role of the softmax outputs as an uncertainty measure. This current study addresses additional hypotheses and reports on the following findings: (i) the performance gain after adding the final BN layer in highly imbalanced settings could still be achieved after removing this additional BN layer in inference; (ii) there is a certain threshold for the imbalance ratio upon which the progress gained by the final BN layer reaches its peak; (iii) the batch size also plays a role and affects the outcome of the final BN application; (iv) the impact of the BN application is also reproducible on other datasets and when utilizing much simpler neural architectures; (v) the reported BN effect occurs only per a single majority class and multiple minority classes i.e., no improvements are evident when there are two majority classes; and finally, (vi) utilizing this BN layer with sigmoid activation has almost no impact when dealing with a strongly imbalanced image classification tasks.

arXiv:2109.09246

Splitfed learning without client-side synchronization: Analyzing client-side split network portion size to overall performance

Federated Learning (FL), Split Learning (SL), and SplitFed Learning (SFL) are three recent developments in distributed machine learning that are gaining attention due to their ability to preserve the privacy of raw data. Thus, they are widely applicable in various domains where data is sensitive, such as large-scale medical image classification, internet-of-medical-things, and cross-organization phishing email detection. SFL is developed on the confluence point of FL and SL. It brings the best of FL and SL by providing parallel client-side machine learning model updates from the FL paradigm and a higher level of model privacy (while training) by splitting the model between the clients and server coming from SL. However, SFL has communication and computation overhead at the client-side due to the requirement of client-side model synchronization. For the resource-constrained client-side, removal of such requirements is required to gain efficiency in the learning. In this regard, this paper studies SFL without client-side model synchronization. The resulting architecture is known as Multi-head Split Learning. Our empirical studies considering the ResNet18 model on MNIST data under IID data distribution among distributed clients find that Multi-head Split Learning is feasible. Its performance is comparable to the SFL. Moreover, SFL provides only 1%-2% better accuracy than Multi-head Split Learning on the MNIST test set. To further strengthen our results, we study the Multi-head Split Learning with various client-side model portions and its impact on the overall performance. To this end, our results find a minimal impact on the overall performance of the model.