PyTorchKR Papers

최신 논문

총 10,163편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:1909.03824

Testing Deep Learning Models for Image Analysis Using Object-Relevant Metamorphic Relations

Deep learning models are widely used for image analysis. While they offer high performance in terms of accuracy, people are concerned about if these models inappropriately make inferences using irrelevant features that are not encoded from the target object in a given image. To address the concern, we propose a metamorphic testing approach that assesses if a given inference is made based on irrelevant features. Specifically, we propose two novel metamorphic relations to detect such inappropriate inferences. We applied our approach to 10 image classification models and 10 object detection models, with three large datasets, i.e., ImageNet, COCO, and Pascal VOC. Over 5.3% of the top-5 correct predictions made by the image classification models are subject to inappropriate inferences using irrelevant features. The corresponding rate for the object detection models is over 8.5%. Based on the findings, we further designed a new image generation strategy that can effectively attack existing models. Comparing with a baseline approach, our strategy can double the success rate of attacks.

arXiv:1909.03388공식 코드 · Papers with Code

L_DMI: An Information-theoretic Noise-robust Loss Function

Accurately annotating large scale dataset is notoriously expensive both in time and in money. Although acquiring low-quality-annotated dataset can be much cheaper, it often badly damages the performance of trained models when using such dataset without particular treatment. Various methods have been proposed for learning with noisy labels. However, most methods only handle limited kinds of noise patterns, require auxiliary information or steps (e.g. , knowing or estimating the noise transition matrix), or lack theoretical justification. In this paper, we propose a novel information-theoretic loss function, $\mathcal{L}_{DMI}$, for training deep neural networks robust to label noise. The core of $\mathcal{L}_{DMI}$ is a generalized version of mutual information, termed Determinant based Mutual Information (DMI), which is not only information-monotone but also relatively invariant. \emph{To the best of our knowledge, $\mathcal{L}_{DMI}$ is the first loss function that is provably robust to instance-independent label noise, regardless of noise pattern, and it can be applied to any existing classification neural networks straightforwardly without any auxiliary information}. In addition to theoretical justification, we also empirically show that using $\mathcal{L}_{DMI}$ outperforms all other counterparts in the classification task on both image dataset and natural language dataset include Fashion-MNIST, CIFAR-10, Dogs vs. Cats, MR with a variety of synthesized noise patterns and noise amounts, as well as a real-world dataset Clothing1M. Codes are available at https://github.com/Newbeeer/L_DMI .

arXiv:1909.03309

Exploring Temporal Differences in 3D Convolutional Neural Networks

Traditional 3D convolutions are computationally expensive, memory intensive, and due to large number of parameters, they often tend to overfit. On the other hand, 2D CNNs are less computationally expensive and less memory intensive than 3D CNNs and have shown remarkable results in applications like image classification and object recognition. However, in previous works, it has been observed that they are inferior to 3D CNNs when applied on a spatio-temporal input. In this work, we propose a convolutional block which extracts the spatial information by performing a 2D convolution and extracts the temporal information by exploiting temporal differences, i.e., the change in the spatial information at different time instances, using simple operations of shift, subtract and add without utilizing any trainable parameters. The proposed convolutional block has same number of parameters as of a 2D convolution kernel of size nxn, i.e. n^2, and has n times lesser parameters than an nxnxn 3D convolution kernel. We show that the 3D CNNs perform better when the 3D convolution kernels are replaced by the proposed convolutional blocks. We evaluate the proposed convolutional block on UCF101 and ModelNet datasets.

arXiv:1909.03585

Learning to Sample: an Active Learning Framework

Meta-learning algorithms for active learning are emerging as a promising paradigm for learning the ``best'' active learning strategy. However, current learning-based active learning approaches still require sufficient training data so as to generalize meta-learning models for active learning. This is contrary to the nature of active learning which typically starts with a small number of labeled samples. The unavailability of large amounts of labeled samples for training meta-learning models would inevitably lead to poor performance (e.g., instabilities and overfitting). In our paper, we tackle these issues by proposing a novel learning-based active learning framework, called Learning To Sample (LTS). This framework has two key components: a sampling model and a boosting model, which can mutually learn from each other in iterations to improve the performance of each other. Within this framework, the sampling model incorporates uncertainty sampling and diversity sampling into a unified process for optimization, enabling us to actively select the most representative and informative samples based on an optimized integration of uncertainty and diversity. To evaluate the effectiveness of the LTS framework, we have conducted extensive experiments on three different classification tasks: image classification, salary level prediction, and entity resolution. The experimental results show that our LTS framework significantly outperforms all the baselines when the label budget is limited, especially for datasets with highly imbalanced classes. In addition to this, our LTS framework can effectively tackle the cold start problem occurring in many existing active learning approaches.

arXiv:1909.03615

Neural Architecture Search in Embedding Space

The neural architecture search (NAS) algorithm with reinforcement learning can be a powerful and novel framework for the automatic discovering process of neural architectures. However, its application is restricted by noncontinuous and high-dimensional search spaces, which result in difficulty in optimization. To resolve these problems, we proposed NAS in embedding space (NASES), which is a novel framework. Unlike other NAS with reinforcement learning approaches that search over a discrete and high-dimensional architecture space, this approach enables reinforcement learning to search in an embedding space by using architecture encoders and decoders. The current experiment demonstrated that the performance of the final architecture network using the NASES procedure is comparable with that of other popular NAS approaches for the image classification task on CIFAR-10. The results of the experiment were efficient and indicated that NASES was highly efficient to discover final architecture only in $<$3.5 GPU hours. The beneficial-performance and effectiveness of NASES was impressive when the architecture-embedding searching and weight initialization were applied.

arXiv:1909.04021

Understanding the Effects of Pre-Training for Object Detectors via Eigenspectrum

ImageNet pre-training has been regarded as essential for training accurate object detectors for a long time. Recently, it has been shown that object detectors trained from randomly initialized weights can be on par with those fine-tuned from ImageNet pre-trained models. However, the effects of pre-training and the differences caused by pre-training are still not fully understood. In this paper, we analyze the eigenspectrum dynamics of the covariance matrix of each feature map in object detectors. Based on our analysis on ResNet-50, Faster R-CNN with FPN, and Mask R-CNN, we show that object detectors trained from ImageNet pre-trained models and those trained from scratch behave differently from each other even if both object detectors have similar accuracy. Furthermore, we propose a method for automatically determining the widths (the numbers of channels) of object detectors based on the eigenspectrum. We train Faster R-CNN with FPN from randomly initialized weights, and show that our method can reduce ~27% of the parameters of ResNet-50 without increasing Multiply-Accumulate operations and losing accuracy. Our results indicate that we should develop more appropriate methods for transferring knowledge from image classification to object detection (or other tasks).

arXiv:1909.02729공식 코드 · Papers with Code

A Baseline for Few-Shot Image Classification

Fine-tuning a deep network trained with the standard cross-entropy loss is a strong baseline for few-shot learning. When fine-tuned transductively, this outperforms the current state-of-the-art on standard datasets such as Mini-ImageNet, Tiered-ImageNet, CIFAR-FS and FC-100 with the same hyper-parameters. The simplicity of this approach enables us to demonstrate the first few-shot learning results on the ImageNet-21k dataset. We find that using a large number of meta-training classes results in high few-shot accuracies even for a large number of few-shot classes. We do not advocate our approach as the solution for few-shot learning, but simply use the results to highlight limitations of current benchmarks and few-shot protocols. We perform extensive studies on benchmark datasets to propose a metric that quantifies the "hardness" of a few-shot episode. This metric can be used to report the performance of few-shot algorithms in a more systematic way.

arXiv:1909.01994

Quasi-Newton Optimization Methods For Deep Learning Applications

Deep learning algorithms often require solving a highly non-linear and nonconvex unconstrained optimization problem. Methods for solving optimization problems in large-scale machine learning, such as deep learning and deep reinforcement learning (RL), are generally restricted to the class of first-order algorithms, like stochastic gradient descent (SGD). While SGD iterates are inexpensive to compute, they have slow theoretical convergence rates. Furthermore, they require exhaustive trial-and-error to fine-tune many learning parameters. Using second-order curvature information to find search directions can help with more robust convergence for non-convex optimization problems. However, computing Hessian matrices for large-scale problems is not computationally practical. Alternatively, quasi-Newton methods construct an approximate of the Hessian matrix to build a quadratic model of the objective function. Quasi-Newton methods, like SGD, require only first-order gradient information, but they can result in superlinear convergence, which makes them attractive alternatives to SGD. The limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) approach is one of the most popular quasi-Newton methods that construct positive definite Hessian approximations. In this chapter, we propose efficient optimization methods based on L-BFGS quasi-Newton methods using line search and trust-region strategies. Our methods bridge the disparity between first- and second-order methods by using gradient information to calculate low-rank updates to Hessian approximations. We provide formal convergence analysis of these methods as well as empirical results on deep learning applications, such as image classification tasks and deep reinforcement learning on a set of ATARI 2600 video games. Our results show a robust convergence with preferred generalization characteristics as well as fast training time.

arXiv:1909.02293

Efficient Neural Architecture Transformation Searchin Channel-Level for Object Detection

Recently, Neural Architecture Search has achieved great success in large-scale image classification. In contrast, there have been limited works focusing on architecture search for object detection, mainly because the costly ImageNet pre-training is always required for detectors. Training from scratch, as a substitute, demands more epochs to converge and brings no computation saving. To overcome this obstacle, we introduce a practical neural architecture transformation search(NATS)algorithm for object detection in this paper. Instead of searching and constructing an entire network, NATS explores the architecture space on the base of existing network and reusing its weights. We propose a novel neural architecture search strategy in channel-level instead of path-level and devise a search space specially targeting at object detection. With the combination of these two designs, an architecture transformation scheme could be discovered to adapt a network designed for image classification to task of object detection. Since our method is gradient-based and only searches for a transformation scheme, the weights of models pretrained inImageNet could be utilized in both searching and retraining stage, which makes the whole process very efficient. The transformed network requires no extra parameters and FLOPs, and is friendly to hardware optimization, which is practical to use in real-time application. In experiments, we demonstrate the effectiveness of NATSon networks like ResNet and ResNeXt. Our transformed networks, combined with various detection frameworks, achieve significant improvements on the COCO dataset while keeping fast.

arXiv:1909.01539

What Happens on the Edge, Stays on the Edge: Toward Compressive Deep Learning

Machine learning at the edge offers great benefits such as increased privacy and security, low latency, and more autonomy. However, a major challenge is that many devices, in particular edge devices, have very limited memory, weak processors, and scarce energy supply. We propose a hybrid hardware-software framework that has the potential to significantly reduce the computational complexity and memory requirements of on-device machine learning. In the first step, inspired by compressive sensing, data is collected in compressed form simultaneously with the sensing process. Thus this compression happens already at the hardware level during data acquisition. But unlike in compressive sensing, this compression is achieved via a projection operator that is specifically tailored to the desired machine learning task. The second step consists of a specially designed and trained deep network. As concrete example we consider the task of image classification, although the proposed framework is more widely applicable. An additional benefit of our approach is that it can be easily combined with existing on-device techniques. Numerical simulations illustrate the viability of our method.

arXiv:1909.00182

Scale Calibrated Training: Improving Generalization of Deep Networks via Scale-Specific Normalization

Standard convolutional neural networks(CNNs) require consistent image resolutions in both training and testing phase. However, in practice, testing with smaller image sizes is necessary for fast inference. We show that trivially evaluating low-resolution images on networks trained with high-resolution images results in a catastrophic accuracy drop in standard CNN architectures. We propose a novel training regime called Scale calibrated Training(SCT) which allows networks to learn from various scales of input simultaneously. By taking advantages of SCT, single network can provide decent accuracy at test time in response to multiple test scales. In our analysis, we surprisingly find that vanilla batch normalization can lead to sub-optimal performance in SCT. Therefore, a novel normalization scheme called Scale-Specific Batch Normalization is equipped to SCT in replacement of batch normalization. Experiment results show that SCT improves accuracy of single Resnet-50 on ImageNet by 1.7% and 11.5% accuracy when testing on image sizes of 224 and 128 respectively.

arXiv:1909.00125

Detecting floodwater on roadways from image data with handcrafted features and deep transfer learning

Detecting roadway segments inundated due to floodwater has important applications for vehicle routing and traffic management decisions. This paper proposes a set of algorithms to automatically detect floodwater that may be present in an image captured by mobile phones or other types of optical cameras. For this purpose, image classification and flood area segmentation methods are developed. For the classification task, we used Local Binary Patterns (LBP), Histogram of Oriented Gradients (HOG) and pre-trained deep neural network (VGG-16) as feature extractors and trained logistic regression, k-nearest neighbors, and decision tree classifiers on the extracted features. Pre-trained VGG-16 network with logistic regression classifier outperformed all other methods. For the flood area segmentation task, we investigated superpixel based methods and Fully Convolutional Neural Network (FCN). Similar to the classification task, we trained logistic regression and k-nearest neighbors classifiers on the superpixel areas and compared that with an end-to-end trained FCN. Conditional Random Fields (CRF) method was applied after both segmentation methods to post-process coarse segmentation results. FCN offered the highest scores in all metrics; it was followed by superpixel-based logistic regression and then superpixel-based KNN.

arXiv:1909.01056

STaDA: Style Transfer as Data Augmentation

The success of training deep Convolutional Neural Networks (CNNs) heavily depends on a significant amount of labelled data. Recent research has found that neural style transfer algorithms can apply the artistic style of one image to another image without changing the latter's high-level semantic content, which makes it feasible to employ neural style transfer as a data augmentation method to add more variation to the training dataset. The contribution of this paper is a thorough evaluation of the effectiveness of the neural style transfer as a data augmentation method for image classification tasks. We explore the state-of-the-art neural style transfer algorithms and apply them as a data augmentation method on Caltech 101 and Caltech 256 dataset, where we found around 2% improvement from 83% to 85% of the image classification accuracy with VGG16, compared with traditional data augmentation strategies. We also combine this new method with conventional data augmentation approaches to further improve the performance of image classification. This work shows the potential of neural style transfer in computer vision field, such as helping us to reduce the difficulty of collecting sufficient labelled data and improve the performance of generic image-based deep learning algorithms.

arXiv:1908.08997

Gradient Weighted Superpixels for Interpretability in CNNs

As Convolutional Neural Networks embed themselves into our everyday lives, the need for them to be interpretable increases. However, there is often a trade-off between methods that are efficient to compute but produce an explanation that is difficult to interpret, and those that are slow to compute but provide a more interpretable result. This is particularly challenging in problem spaces that require a large input volume, especially video which combines both spatial and temporal dimensions. In this work we introduce the idea of scoring superpixels through the use of gradient based pixel scoring techniques. We show qualitatively and quantitatively that this is able to approximate LIME, in a fraction of the time. We investigate our techniques using both image classification, and action recognition networks on large scale datasets (ImageNet and Kinetics-400 respectively).

arXiv:1908.08988공식 코드 · Papers with Code

Neural Image Compression and Explanation

Explaining the prediction of deep neural networks (DNNs) and semantic image compression are two active research areas of deep learning with a numerous of applications in decision-critical systems, such as surveillance cameras, drones and self-driving cars, where interpretable decision is critical and storage/network bandwidth is limited. In this paper, we propose a novel end-to-end Neural Image Compression and Explanation (NICE) framework that learns to (1) explain the predictions of convolutional neural networks (CNNs), and (2) subsequently compress the input images for efficient storage or transmission. Specifically, NICE generates a sparse mask over an input image by attaching a stochastic binary gate to each pixel of the image, whose parameters are learned through the interaction with the CNN classifier to be explained. The generated mask is able to capture the saliency of each pixel measured by its influence to the final prediction of CNN; it can also be used to produce a mixed-resolution image, where important pixels maintain their original high resolution and insignificant background pixels are subsampled to a low resolution. The produced images achieve a high compression rate (e.g., about 0.6x of original image file size), while retaining a similar classification accuracy. Extensive experiments across multiple image classification benchmarks demonstrate the superior performance of NICE compared to the state-of-the-art methods in terms of explanation quality and semantic image compression rate. Our code is available at: https://github.com/lxuniverse/NICE.

arXiv:1908.08987

PCGAN-CHAR: Progressively Trained Classifier Generative Adversarial Networks for Classification of Noisy Handwritten Bangla Characters

Due to the sparsity of features, noise has proven to be a great inhibitor in the classification of handwritten characters. To combat this, most techniques perform denoising of the data before classification. In this paper, we consolidate the approach by training an all-in-one model that is able to classify even noisy characters. For classification, we progressively train a classifier generative adversarial network on the characters from low to high resolution. We show that by learning the features at each resolution independently a trained model is able to accurately classify characters even in the presence of noise. We experimentally demonstrate the effectiveness of our approach by classifying noisy versions of MNIST, handwritten Bangla Numeral, and Basic Character datasets.

arXiv:1908.08984공식 코드 · Papers with Code

Atlas: A Dataset and Benchmark for E-commerce Clothing Product Categorization

In E-commerce, it is a common practice to organize the product catalog using product taxonomy. This enables the buyer to easily locate the item they are looking for and also to explore various items available under a category. Product taxonomy is a tree structure with 3 or more levels of depth and several leaf nodes. Product categorization is a large scale classification task that assigns a category path to a particular product. Research in this area is restricted by the unavailability of good real-world datasets and the variations in taxonomy due to the absence of a standard across the different e-commerce stores. In this paper, we introduce a high-quality product taxonomy dataset focusing on clothing products which contain 186,150 images under clothing category with 3 levels and 52 leaf nodes in the taxonomy. We explain the methodology used to collect and label this dataset. Further, we establish the benchmark by comparing image classification and Attention based Sequence models for predicting the category path. Our benchmark model reaches a micro f-score of 0.92 on the test set. The dataset, code and pre-trained models are publicly available at \url{https://github.com/vumaasha/atlas}. We invite the community to improve upon these baselines.

arXiv:1908.11645공식 코드 · Papers with Code

EBPC: Extended Bit-Plane Compression for Deep Neural Network Inference and Training Accelerators

In the wake of the success of convolutional neural networks in image classification, object recognition, speech recognition, etc., the demand for deploying these compute-intensive ML models on embedded and mobile systems with tight power and energy constraints at low cost, as well as for boosting throughput in data centers, is growing rapidly. This has sparked a surge of research into specialized hardware accelerators. Their performance is typically limited by I/O bandwidth, power consumption is dominated by I/O transfers to off-chip memory, and on-chip memories occupy a large part of the silicon area. We introduce and evaluate a novel, hardware-friendly, and lossless compression scheme for the feature maps present within convolutional neural networks. We present hardware architectures and synthesis results for the compressor and decompressor in 65nm. With a throughput of one 8-bit word/cycle at 600MHz, they fit into 2.8kGE and 3.0kGE of silicon area, respectively - together the size of less than seven 8-bit multiply-add units at the same throughput. We show that an average compression ratio of 5.1x for AlexNet, 4x for VGG-16, 2.4x for ResNet-34 and 2.2x for MobileNetV2 can be achieved - a gain of 45-70% over existing methods. Our approach also works effectively for various number formats, has a low frame-to-frame variance on the compression ratio, and achieves compression factors for gradient map compression during training that are even better than for inference.

arXiv:1908.11528

Bin-wise Temperature Scaling (BTS): Improvement in Confidence Calibration Performance through Simple Scaling Techniques

The prediction reliability of neural networks is important in many applications. Specifically, in safety-critical domains, such as cancer prediction or autonomous driving, a reliable confidence of model's prediction is critical for the interpretation of the results. Modern deep neural networks have achieved a significant improvement in performance for many different image classification tasks. However, these networks tend to be poorly calibrated in terms of output confidence. Temperature scaling is an efficient post-processing-based calibration scheme and obtains well calibrated results. In this study, we leverage the concept of temperature scaling to build a sophisticated bin-wise scaling. Furthermore, we adopt augmentation of validation samples for elaborated scaling. The proposed methods consistently improve calibration performance with various datasets and deep convolutional neural network models.

논문 ID: document/8489929

Which Has Better Visual Quality: The Clear Blue Sky or a Blurry Animal?

Image content variation is a typical and challenging problem in no-reference image quality assessment (NR-IQA). This work pays special attention to the impact of image content variation on NR-IQA methods. To better analyze this impact, we focus on blur-dominated distortions to exclude the impacts of distortion-type variations. We empirically show that current NR-IQA methods are inconsistent with human visual perception when predicting the relative quality of image pairs with different image contents. In view of deep semantic features of pretrained image classification neural networks always containing discriminative image content information, we put forward a new NR-IQA method based on semantic feature aggregation (SFA) to alleviate the impact of image content variation. Specifically, instead of resizing the image, we first crop multiple overlapping patches over the entire distorted image to avoid introducing geometric deformations. Then, according to an adaptive layer selection procedure, we extract deep semantic features by leveraging the power of a pretrained image classification model for its inherent content-aware property. After that, the local patch features are aggregated using several statistical structures. Finally, a linear regression model is trained for mapping the aggregated global features to image-quality scores. The proposed method, SFA, is compared with nine representative blur-specific NR-IQA methods, two general-purpose NR-IQA methods, and two extra full-reference IQA methods on Gaussian blur images (with and without Gaussian noise/JPEG compression) and realistic blur images from multiple databases, including LIVE, TID2008, TID2013, MLIVE1, MLIVE2, BID, and CLIVE. Experimental results show that SFA is superior to the state-of-the-art NR methods on all seven databases. It is also verified that deep semantic features play a crucial role in addressing image content variation, and this provides a new perspective for NR-IQA.