PyTorchKR Papers

최신 논문

총 5,575편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:1802.08435

Efficient Neural Audio Synthesis

Sequential models achieve state-of-the-art results in audio, visual and textual domains with respect to both estimating the data distribution and generating high-quality samples. Efficient sampling for this class of models has however remained an elusive problem. With a focus on text-to-speech synthesis, we describe a set of general techniques for reducing sampling time while maintaining high output quality. We first describe a single-layer recurrent neural network, the WaveRNN, with a dual softmax layer that matches the quality of the state-of-the-art WaveNet model. The compact form of the network makes it possible to generate 24kHz 16-bit audio 4x faster than real time on a GPU. Second, we apply a weight pruning technique to reduce the number of weights in the WaveRNN. We find that, for a constant number of parameters, large sparse networks perform better than small dense networks and this relationship holds for sparsity levels beyond 96%. The small number of weights in a Sparse WaveRNN makes it possible to sample high-fidelity audio on a mobile CPU in real time. Finally, we propose a new generation scheme based on subscaling that folds a long sequence into a batch of shorter sequences and allows one to generate multiple samples at once. The Subscale WaveRNN produces 16 samples per step without loss of quality and offers an orthogonal method for increasing sampling efficiency.

arXiv:1802.08362

Real-Time End-to-End Action Detection with Two-Stream Networks

Two-stream networks have been very successful for solving the problem of action detection. However, prior work using two-stream networks train both streams separately, which prevents the network from exploiting regularities between the two streams. Moreover, unlike the visual stream, the dominant forms of optical flow computation typically do not maximally exploit GPU parallelism. We present a real-time end-to-end trainable two-stream network for action detection. First, we integrate the optical flow computation in our framework by using Flownet2. Second, we apply early fusion for the two streams and train the whole pipeline jointly end-to-end. Finally, for better network initialization, we transfer from the task of action recognition to action detection by pre-training our framework using the recently released large-scale Kinetics dataset. Our experimental results show that training the pipeline jointly end-to-end with fine-tuning the optical flow for the objective of action detection improves detection performance significantly. Additionally, we observe an improvement when initializing with parameters pre-trained using Kinetics. Last, we show that by integrating the optical flow computation, our framework is more efficient, running at real-time speeds (up to 31 fps).

arXiv:1802.08800

Stochastic Gradient Descent on Highly-Parallel Architectures

There is an increased interest in building data analytics frameworks with advanced algebraic capabilities both in industry and academia. Many of these frameworks, e.g., TensorFlow and BIDMach, implement their compute-intensive primitives in two flavors---as multi-thread routines for multi-core CPUs and as highly-parallel kernels executed on GPU. Stochastic gradient descent (SGD) is the most popular optimization method for model training implemented extensively on modern data analytics platforms. While the data-intensive properties of SGD are well-known, there is an intense debate on which of the many SGD variants is better in practice. In this paper, we perform a comprehensive study of parallel SGD for training generalized linear models. We consider the impact of three factors -- computing architecture (multi-core CPU or GPU), synchronous or asynchronous model updates, and data sparsity -- on three measures---hardware efficiency, statistical efficiency, and time to convergence. In the process, we design an optimized asynchronous SGD algorithm for GPU that leverages warp shuffling and cache coalescing for data and model access. We draw several interesting findings from our extensive experiments with logistic regression (LR) and support vector machines (SVM) on five real datasets. For synchronous SGD, GPU always outperforms parallel CPU---they both outperform a sequential CPU solution by more than 400X. For asynchronous SGD, parallel CPU is the safest choice while GPU with data replication is better in certain situations. The choice between synchronous GPU and asynchronous CPU depends on the task and the characteristics of the data. As a reference, our best implementation outperforms TensorFlow and BIDMach consistently. We hope that our insights provide a useful guide for applying parallel SGD to generalized linear models.

arXiv:1603.08390공식 코드 · Papers with Code

A Generic Inverted Index Framework for Similarity Search on the GPU - Technical Report

We propose a novel generic inverted index framework on the GPU (called GENIE), aiming to reduce the programming complexity of the GPU for parallel similarity search of different data types. Not every data type and similarity measure are supported by GENIE, but many popular ones are. We present the system design of GENIE, and demonstrate similarity search with GENIE on several data types along with a theoretical analysis of search results. A new concept of locality sensitive hashing (LSH) named $\tau$-ANN search, and a novel data structure c-PQ on the GPU are also proposed for achieving this purpose. Extensive experiments on different real-life datasets demonstrate the efficiency and effectiveness of our framework. The implemented system has been released as open source.

arXiv:1803.00737

Fusion of multispectral satellite imagery using a cluster of graphics processing unit

The paper presents a parallel implementation of existing image fusion methods on a graphical cluster. Parallel implementations of methods based on discrete wavelet transformation (Haars and Daubechies discrete wavelet transform) are developed. Experiments were performed on a cluster using GPU and CPU and performance gains were estimated for the use of the developed parallel implementations to process satellite images from satellite Landsat 7. The implementation on a graphic cluster provides performance improvement from 2 to 18 times. The quality of the considered methods was evaluated by ERGAS and QNR metrics. The results show performance gains and retaining of quality with the cluster of GPU compared to the results obtained by the authors and other researchers for a CPU and single GPU.

arXiv:1803.01516

A new stereo formulation not using pixel and disparity models

We introduce a new stereo formulation which does not use pixel and disparity models. Many problems in vision are treated as assigning each pixel a label. Disparities are labels for stereo. Such pixel-labeling problems are naturally represented in terms of energy minimization, where the energy function has two terms: one term penalizes solutions that inconsistent with the observed data, the other term enforces spatial smoothness. Graph cuts are one of the effi- cient methods for solving energy minimization. However, exact minimization of multi labeling problems can be performed by graph cuts only for the case with convex smoothness terms. In pixel-disparity formulation, convex smoothness terms do not generate well reconstructed 3D results. Thus, truncated linear or quadratic smoothness terms, etc. are used, where approximate energy minimization is necessary. In this paper, we introduce a new site-labeling formulation, where the sites are not pixels but lines in 3D space, labels are not disparities but depth numbers. For this formulation, visibility reasoning is naturally included in the energy function. In addition, this formulation allows us to use a small smoothness term, which does not affect the 3D results much. This makes the optimization step very simple, so we could develop an approximation method for graph cut itself (not for energy minimization) and a high performance GPU graph cut program. For Tsukuba stereo pair in Middlebury data set, we got the result in 5ms using GTX1080GPU, 19ms using GTX660GPU.

arXiv:1802.09113

GPU Accelerated Sub-Sampled Newton's Method

First order methods, which solely rely on gradient information, are commonly used in diverse machine learning (ML) and data analysis (DA) applications. This is attributed to the simplicity of their implementations, as well as low per-iteration computational/storage costs. However, they suffer from significant disadvantages; most notably, their performance degrades with increasing problem ill-conditioning. Furthermore, they often involve a large number of hyper-parameters, and are notoriously sensitive to parameters such as the step-size. By incorporating additional information from the Hessian, second-order methods, have been shown to be resilient to many such adversarial effects. However, these advantages of using curvature information come at the cost of higher per-iteration costs, which in \enquote{big data} regimes, can be computationally prohibitive. In this paper, we show that, contrary to conventional belief, second-order methods, when implemented appropriately, can be more efficient than first-order alternatives in many large-scale ML/ DA applications. In particular, in convex settings, we consider variants of classical Newton\textsf{'}s method in which the Hessian and/or the gradient are randomly sub-sampled. We show that by effectively leveraging the power of GPUs, such randomized Newton-type algorithms can be significantly accelerated, and can easily outperform state of the art implementations of existing techniques in popular ML/ DA software packages such as TensorFlow. Additionally these randomized methods incur a small memory overhead compared to first-order methods. In particular, we show that for million-dimensional problems, our GPU accelerated sub-sampled Newton\textsf{'}s method achieves a higher test accuracy in milliseconds as compared with tens of seconds for first order alternatives.

arXiv:1710.08961

Fast and Scalable Distributed Deep Convolutional Autoencoder for fMRI Big Data Analytics

In recent years, analyzing task-based fMRI (tfMRI) data has become an essential tool for understanding brain function and networks. However, due to the sheer size of tfMRI data, its intrinsic complex structure, and lack of ground truth of underlying neural activities, modeling tfMRI data is hard and challenging. Previously proposed data-modeling methods including Independent Component Analysis (ICA) and Sparse Dictionary Learning only provided a weakly established model based on blind source separation under the strong assumption that original fMRI signals could be linearly decomposed into time series components with corresponding spatial maps. Meanwhile, analyzing and learning a large amount of tfMRI data from a variety of subjects has been shown to be very demanding but yet challenging even with technological advances in computational hardware. Given the Convolutional Neural Network (CNN), a robust method for learning high-level abstractions from low-level data such as tfMRI time series, in this work we propose a fast and scalable novel framework for distributed deep Convolutional Autoencoder model. This model aims to both learn the complex hierarchical structure of the tfMRI data and to leverage the processing power of multiple GPUs in a distributed fashion. To implement such a model, we have created an enhanced processing pipeline on the top of Apache Spark and Tensorflow library, leveraging from a very large cluster of GPU machines. Experimental data from applying the model on the Human Connectome Project (HCP) show that the proposed model is efficient and scalable toward tfMRI big data analytics, thus enabling data-driven extraction of hierarchical neuroscientific information from massive fMRI big data in the future.

arXiv:1803.02699

CNN-Based Automatic Urinary Particles Recognition

The urine sediment analysis of particles in microscopic images can assist physicians in evaluating patients with renal and urinary tract diseases. Manual urine sediment examination is labor-intensive, subjective and time-consuming, and the traditional automatic algorithms often extract the hand-crafted features for recognition. Instead of using the hand-crafted features, in this paper, we exploit CNN to learn features in an end-to-end manner to recognize the urine particles. We treat the urine particles recognition as object detection and exploit two state-of-the-art CNN-based object detection methods, Faster R-CNN and SSD, as well as their variants for urine particles recognition. We further investigate different factors involving these CNN-based object detection methods for urine particles recognition. We comprehensively evaluate these methods on a dataset consisting of 5,376 annotated images corresponding to 7 categories of urine particles, i.e., erythrocyte, leukocyte, epithelial cell, crystal, cast, mycete, epithelial nuclei, and obtain a best mAP (mean average precision) of 84.1% while taking only 72 ms per image on a NVIDIA Titan X GPU.

arXiv:1803.02181

2^B3^C: 2 Box 3 Crop of Facial Image for Gender Classification with Convolutional Networks

In this paper, we tackle the classification of gender in facial images with deep learning. Our convolutional neural networks (CNN) use the VGG-16 architecture [1] and are pretrained on ImageNet for image classification. Our proposed method (2^B3^C) first detects the face in the facial image, increases the margin of a detected face by 50%, cropping the face with two boxes three crop schemes (Left, Middle, and Right crop) and extracts the CNN predictions on the cropped schemes. The CNNs of our method is fine-tuned on the Adience and LFW with gender annotations. We show the effectiveness of our method by achieving 90.8% classification on Adience and achieving competitive 95.3% classification accuracy on LFW dataset. In addition, to check the true ability of our method, our gender classification system has a frame rate of 7-10 fps (frames per seconds) on a GPU considering real-time scenarios.

arXiv:1705.07175공식 코드 · Papers with Code

Espresso: Efficient Forward Propagation for BCNNs

There are many applications scenarios for which the computational performance and memory footprint of the prediction phase of Deep Neural Networks (DNNs) needs to be optimized. Binary Neural Networks (BDNNs) have been shown to be an effective way of achieving this objective. In this paper, we show how Convolutional Neural Networks (CNNs) can be implemented using binary representations. Espresso is a compact, yet powerful library written in C/CUDA that features all the functionalities required for the forward propagation of CNNs, in a binary file less than 400KB, without any external dependencies. Although it is mainly designed to take advantage of massive GPU parallelism, Espresso also provides an equivalent CPU implementation for CNNs. Espresso provides special convolutional and dense layers for BCNNs, leveraging bit-packing and bit-wise computations for efficient execution. These techniques provide a speed-up of matrix-multiplication routines, and at the same time, reduce memory usage when storing parameters and activations. We experimentally show that Espresso is significantly faster than existing implementations of optimized binary neural networks ($\approx$ 2 orders of magnitude). Espresso is released under the Apache 2.0 license and is available at http://github.com/fpeder/espresso.

arXiv:1803.03778

Driving Scene Perception Network: Real-time Joint Detection, Depth Estimation and Semantic Segmentation

As the demand for enabling high-level autonomous driving has increased in recent years and visual perception is one of the critical features to enable fully autonomous driving, in this paper, we introduce an efficient approach for simultaneous object detection, depth estimation and pixel-level semantic segmentation using a shared convolutional architecture. The proposed network model, which we named Driving Scene Perception Network (DSPNet), uses multi-level feature maps and multi-task learning to improve the accuracy and efficiency of object detection, depth estimation and image segmentation tasks from a single input image. Hence, the resulting network model uses less than 850 MiB of GPU memory and achieves 14.0 fps on NVIDIA GeForce GTX 1080 with a 1024x512 input image, and both precision and efficiency have been improved over combination of single tasks.

arXiv:1803.05675공식 코드 · Papers with Code

Training of Convolutional Networks on Multiple Heterogeneous Datasets for Street Scene Semantic Segmentation

We propose a convolutional network with hierarchical classifiers for per-pixel semantic segmentation, which is able to be trained on multiple, heterogeneous datasets and exploit their semantic hierarchy. Our network is the first to be simultaneously trained on three different datasets from the intelligent vehicles domain, i.e. Cityscapes, GTSDB and Mapillary Vistas, and is able to handle different semantic level-of-detail, class imbalances, and different annotation types, i.e. dense per-pixel and sparse bounding-box labels. We assess our hierarchical approach, by comparing against flat, non-hierarchical classifiers and we show improvements in mean pixel accuracy of 13.0% for Cityscapes classes and 2.4% for Vistas classes and 32.3% for GTSDB classes. Our implementation achieves inference rates of 17 fps at a resolution of 520x706 for 108 classes running on a GPU.

arXiv:1803.05959

Mo2Cap2: Real-time Mobile 3D Motion Capture with a Cap-mounted Fisheye Camera

We propose the first real-time approach for the egocentric estimation of 3D human body pose in a wide range of unconstrained everyday activities. This setting has a unique set of challenges, such as mobility of the hardware setup, and robustness to long capture sessions with fast recovery from tracking failures. We tackle these challenges based on a novel lightweight setup that converts a standard baseball cap to a device for high-quality pose estimation based on a single cap-mounted fisheye camera. From the captured egocentric live stream, our CNN based 3D pose estimation approach runs at 60Hz on a consumer-level GPU. In addition to the novel hardware setup, our other main contributions are: 1) a large ground truth training corpus of top-down fisheye images and 2) a novel disentangled 3D pose estimation approach that takes the unique properties of the egocentric viewpoint into account. As shown by our evaluation, we achieve lower 3D joint error as well as better 2D overlay than the existing baselines.

arXiv:1803.05880

GossipGraD: Scalable Deep Learning using Gossip Communication based Asynchronous Gradient Descent

In this paper, we present GossipGraD - a gossip communication protocol based Stochastic Gradient Descent (SGD) algorithm for scaling Deep Learning (DL) algorithms on large-scale systems. The salient features of GossipGraD are: 1) reduction in overall communication complexity from {\Theta}(log(p)) for p compute nodes in well-studied SGD to O(1), 2) model diffusion such that compute nodes exchange their updates (gradients) indirectly after every log(p) steps, 3) rotation of communication partners for facilitating direct diffusion of gradients, 4) asynchronous distributed shuffle of samples during the feedforward phase in SGD to prevent over-fitting, 5) asynchronous communication of gradients for further reducing the communication cost of SGD and GossipGraD. We implement GossipGraD for GPU and CPU clusters and use NVIDIA GPUs (Pascal P100) connected with InfiniBand, and Intel Knights Landing (KNL) connected with Aries network. We evaluate GossipGraD using well-studied dataset ImageNet-1K (~250GB), and widely studied neural network topologies such as GoogLeNet and ResNet50 (current winner of ImageNet Large Scale Visualization Research Challenge (ILSVRC)). Our performance evaluation using both KNL and Pascal GPUs indicates that GossipGraD can achieve perfect efficiency for these datasets and their associated neural network topologies. Specifically, for ResNet50, GossipGraD is able to achieve ~100% compute efficiency using 128 NVIDIA Pascal P100 GPUs - while matching the top-1 classification accuracy published in literature.

arXiv:1710.03811

DeepSolarEye: Power Loss Prediction and Weakly Supervised Soiling Localization via Fully Convolutional Networks for Solar Panels

The impact of soiling on solar panels is an important and well-studied problem in renewable energy sector. In this paper, we present the first convolutional neural network (CNN) based approach for solar panel soiling and defect analysis. Our approach takes an RGB image of solar panel and environmental factors as inputs to predict power loss, soiling localization, and soiling type. In computer vision, localization is a complex task which typically requires manually labeled training data such as bounding boxes or segmentation masks. Our proposed approach consists of specialized four stages which completely avoids localization ground truth and only needs panel images with power loss labels for training. The region of impact area obtained from the predicted localization masks are classified into soiling types using the webly supervised learning. For improving localization capabilities of CNNs, we introduce a novel bi-directional input-aware fusion (BiDIAF) block that reinforces the input at different levels of CNN to learn input-specific feature maps. Our empirical study shows that BiDIAF improves the power loss prediction accuracy by about 3% and localization accuracy by about 4%. Our end-to-end model yields further improvement of about 24% on localization when learned in a weakly supervised manner. Our approach is generalizable and showed promising results on web crawled solar panel images. Our system has a frame rate of 22 fps (including all steps) on a NVIDIA TitanX GPU. Additionally, we collected first of it's kind dataset for solar panel image analysis consisting 45,000+ images.

arXiv:1803.07289

Flex-Convolution (Million-Scale Point-Cloud Learning Beyond Grid-Worlds)

Traditional convolution layers are specifically designed to exploit the natural data representation of images -- a fixed and regular grid. However, unstructured data like 3D point clouds containing irregular neighborhoods constantly breaks the grid-based data assumption. Therefore applying best-practices and design choices from 2D-image learning methods towards processing point clouds are not readily possible. In this work, we introduce a natural generalization flex-convolution of the conventional convolution layer along with an efficient GPU implementation. We demonstrate competitive performance on rather small benchmark sets using fewer parameters and lower memory consumption and obtain significant improvements on a million-scale real-world dataset. Ours is the first which allows to efficiently process 7 million points concurrently.

arXiv:1803.07950공식 코드 · Papers with Code

End-to-End Video Captioning with Multitask Reinforcement Learning

Although end-to-end (E2E) learning has led to impressive progress on a variety of visual understanding tasks, it is often impeded by hardware constraints (e.g., GPU memory) and is prone to overfitting. When it comes to video captioning, one of the most challenging benchmark tasks in computer vision, those limitations of E2E learning are especially amplified by the fact that both the input videos and output captions are lengthy sequences. Indeed, state-of-the-art methods for video captioning process video frames by convolutional neural networks and generate captions by unrolling recurrent neural networks. If we connect them in an E2E manner, the resulting model is both memory-consuming and data-hungry, making it extremely hard to train. In this paper, we propose a multitask reinforcement learning approach to training an E2E video captioning model. The main idea is to mine and construct as many effective tasks (e.g., attributes, rewards, and the captions) as possible from the human captioned videos such that they can jointly regulate the search space of the E2E neural network, from which an E2E video captioning model can be found and generalized to the testing phase. To the best of our knowledge, this is the first video captioning model that is trained end-to-end from the raw video input to the caption output. Experimental results show that such a model outperforms existing ones to a large margin on two benchmark video captioning datasets.

arXiv:1803.07724공식 코드 · Papers with Code

Attention on Attention: Architectures for Visual Question Answering (VQA)

Visual Question Answering (VQA) is an increasingly popular topic in deep learning research, requiring coordination of natural language processing and computer vision modules into a single architecture. We build upon the model which placed first in the VQA Challenge by developing thirteen new attention mechanisms and introducing a simplified classifier. We performed 300 GPU hours of extensive hyperparameter and architecture searches and were able to achieve an evaluation score of 64.78%, outperforming the existing state-of-the-art single model's validation score of 63.15%.

arXiv:1803.06815공식 코드 · Papers with Code

ESPNet: Efficient Spatial Pyramid of Dilated Convolutions for Semantic Segmentation

We introduce a fast and efficient convolutional neural network, ESPNet, for semantic segmentation of high resolution images under resource constraints. ESPNet is based on a new convolutional module, efficient spatial pyramid (ESP), which is efficient in terms of computation, memory, and power. ESPNet is 22 times faster (on a standard GPU) and 180 times smaller than the state-of-the-art semantic segmentation network PSPNet, while its category-wise accuracy is only 8% less. We evaluated ESPNet on a variety of semantic segmentation datasets including Cityscapes, PASCAL VOC, and a breast biopsy whole slide image dataset. Under the same constraints on memory and computation, ESPNet outperforms all the current efficient CNN networks such as MobileNet, ShuffleNet, and ENet on both standard metrics and our newly introduced performance metrics that measure efficiency on edge devices. Our network can process high resolution images at a rate of 112 and 9 frames per second on a standard GPU and edge device, respectively.