PyTorchKR Papers

최신 논문

총 9,642편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2003.11963공식 코드 · Papers with Code

TLDR: Token Loss Dynamic Reweighting for Reducing Repetitive Utterance Generation

Natural Language Generation (NLG) models are prone to generating repetitive utterances. In this work, we study the repetition problem for encoder-decoder models, using both recurrent neural network (RNN) and transformer architectures. To this end, we consider the chit-chat task, where the problem is more prominent than in other tasks that need encoder-decoder architectures. We first study the influence of model architectures. By using pre-attention and highway connections for RNNs, we manage to achieve lower repetition rates. However, this method does not generalize to other models such as transformers. We hypothesize that the deeper reason is that in the training corpora, there are hard tokens that are more difficult for a generative model to learn than others and, once learning has finished, hard tokens are still under-learned, so that repetitive generations are more likely to happen. Based on this hypothesis, we propose token loss dynamic reweighting (TLDR) that applies differentiable weights to individual token losses. By using higher weights for hard tokens and lower weights for easy tokens, NLG models are able to learn individual tokens at different paces. Experiments on chit-chat benchmark datasets show that TLDR is more effective in repetition reduction for both RNN and transformer architectures than baselines using different weighting functions.

arXiv:2003.11282

Content Adaptive and Error Propagation Aware Deep Video Compression

Recently, learning based video compression methods attract increasing attention. However, the previous works suffer from error propagation due to the accumulation of reconstructed error in inter predictive coding. Meanwhile, the previous learning based video codecs are also not adaptive to different video contents. To address these two problems, we propose a content adaptive and error propagation aware video compression system. Specifically, our method employs a joint training strategy by considering the compression performance of multiple consecutive frames instead of a single frame. Based on the learned long-term temporal information, our approach effectively alleviates error propagation in reconstructed frames. More importantly, instead of using the hand-crafted coding modes in the traditional compression systems, we design an online encoder updating scheme in our system. The proposed approach updates the parameters for encoder according to the rate-distortion criterion but keeps the decoder unchanged in the inference stage. Therefore, the encoder is adaptive to different video contents and achieves better compression performance by reducing the domain gap between the training and testing datasets. Our method is simple yet effective and outperforms the state-of-the-art learning based video codecs on benchmark datasets without increasing the model size or decreasing the decoding speed.

arXiv:2003.11213

A New Multiple Max-pooling Integration Module and Cross Multiscale Deconvolution Network Based on Image Semantic Segmentation

To better retain the deep features of an image and solve the sparsity problem of the end-to-end segmentation model, we propose a new deep convolutional network model for medical image pixel segmentation, called MC-Net. The core of this network model consists of four parts, namely, an encoder network, a multiple max-pooling integration module, a cross multiscale deconvolution decoder network and a pixel-level classification layer. In the network structure of the encoder, we use multiscale convolution instead of the traditional single-channel convolution. The multiple max-pooling integration module first integrates the output features of each submodule of the encoder network and reduces the number of parameters by convolution using a kernel size of 1. At the same time, each max-pooling layer (the pooling size of each layer is different) is spliced after each convolution to achieve the translation invariance of the feature maps of each submodule. We use the output feature maps from the multiple max-pooling integration module as the input of the decoder network; the multiscale convolution of each submodule in the decoder network is cross-fused with the feature maps generated by the corresponding multiscale convolution in the encoder network. Using the above feature map processing methods solves the sparsity problem after the max-pooling layer-generating matrix and enhances the robustness of the classification. We compare our proposed model with the well-known Fully Convolutional Networks for Semantic Segmentation (FCNs), DecovNet, PSPNet, U-net, SgeNet and other state-of-the-art segmentation networks such as HyperDenseNet, MS-Dual, Espnetv2, Denseaspp using one binary Kaggle 2018 data science bowl dataset and two multiclass dataset and obtain encouraging experimental results.

arXiv:2003.11173

Learning Syntactic and Dynamic Selective Encoding for Document Summarization

Text summarization aims to generate a headline or a short summary consisting of the major information of the source text. Recent studies employ the sequence-to-sequence framework to encode the input with a neural network and generate abstractive summary. However, most studies feed the encoder with the semantic word embedding but ignore the syntactic information of the text. Further, although previous studies proposed the selective gate to control the information flow from the encoder to the decoder, it is static during the decoding and cannot differentiate the information based on the decoder states. In this paper, we propose a novel neural architecture for document summarization. Our approach has the following contributions: first, we incorporate syntactic information such as constituency parsing trees into the encoding sequence to learn both the semantic and syntactic information from the document, resulting in more accurate summary; second, we propose a dynamic gate network to select the salient information based on the context of the decoder state, which is essential to document summarization. The proposed model has been evaluated on CNN/Daily Mail summarization datasets and the experimental results show that the proposed approach outperforms baseline approaches.

arXiv:2003.09586

Probing Word Translations in the Transformer and Trading Decoder for Encoder Layers

Due to its effectiveness and performance, the Transformer translation model has attracted wide attention, most recently in terms of probing-based approaches. Previous work focuses on using or probing source linguistic features in the encoder. To date, the way word translation evolves in Transformer layers has not yet been investigated. Naively, one might assume that encoder layers capture source information while decoder layers translate. In this work, we show that this is not quite the case: translation already happens progressively in encoder layers and even in the input embeddings. More surprisingly, we find that some of the lower decoder layers do not actually do that much decoding. We show all of this in terms of a probing approach where we project representations of the layer analyzed to the final trained and frozen classifier level of the Transformer decoder to measure word translation accuracy. Our findings motivate and explain a Transformer configuration change: if translation already happens in the encoder layers, perhaps we can increase the number of encoder layers, while decreasing the number of decoder layers, boosting decoding speed, without loss in translation quality? Our experiments show that this is indeed the case: we can increase speed by up to a factor 2.3 with small gains in translation quality, while an 18-4 deep encoder configuration boosts translation quality by +1.42 BLEU (En-De) at a speed-up of 1.4.

arXiv:2003.09691

Cross-modal Deep Face Normals with Deactivable Skip Connections

We present an approach for estimating surface normals from in-the-wild color images of faces. While data-driven strategies have been proposed for single face images, limited available ground truth data makes this problem difficult. To alleviate this issue, we propose a method that can leverage all available image and normal data, whether paired or not, thanks to a novel cross-modal learning architecture. In particular, we enable additional training with single modality data, either color or normal, by using two encoder-decoder networks with a shared latent space. The proposed architecture also enables face details to be transferred between the image and normal domains, given paired data, through skip connections between the image encoder and normal decoder. Core to our approach is a novel module that we call deactivable skip connections, which allows integrating both the auto-encoded and image-to-normal branches within the same architecture that can be trained end-to-end. This allows learning of a rich latent space that can accurately capture the normal information. We compare against state-of-the-art methods and show that our approach can achieve significant improvements, both quantitative and qualitative, with natural face images.

arXiv:2003.10144

CF2-Net: Coarse-to-Fine Fusion Convolutional Network for Breast Ultrasound Image Segmentation

Breast ultrasound (BUS) image segmentation plays a crucial role in a computer-aided diagnosis system, which is regarded as a useful tool to help increase the accuracy of breast cancer diagnosis. Recently, many deep learning methods have been developed for segmentation of BUS image and show some advantages compared with conventional region-, model-, and traditional learning-based methods. However, previous deep learning methods typically use skip-connection to concatenate the encoder and decoder, which might not make full fusion of coarse-to-fine features from encoder and decoder. Since the structure and edge of lesion in BUS image are common blurred, these would make it difficult to learn the discriminant information of structure and edge, and reduce the performance. To this end, we propose and evaluate a coarse-to-fine fusion convolutional network (CF2-Net) based on a novel feature integration strategy (forming an 'E'-like type) for BUS image segmentation. To enhance contour and provide structural information, we concatenate a super-pixel image and the original image as the input of CF2-Net. Meanwhile, to highlight the differences in the lesion regions with variable sizes and relieve the imbalance issue, we further design a weighted-balanced loss function to train the CF2-Net effectively. The proposed CF2-Net was evaluated on an open dataset by using four-fold cross validation. The results of the experiment demonstrate that the CF2-Net obtains state-of-the-art performance when compared with other deep learning-based methods

arXiv:2003.09005공식 코드 · Papers with Code

Semi-Supervised Semantic Segmentation with Cross-Consistency Training

In this paper, we present a novel cross-consistency based semi-supervised approach for semantic segmentation. Consistency training has proven to be a powerful semi-supervised learning framework for leveraging unlabeled data under the cluster assumption, in which the decision boundary should lie in low-density regions. In this work, we first observe that for semantic segmentation, the low-density regions are more apparent within the hidden representations than within the inputs. We thus propose cross-consistency training, where an invariance of the predictions is enforced over different perturbations applied to the outputs of the encoder. Concretely, a shared encoder and a main decoder are trained in a supervised manner using the available labeled examples. To leverage the unlabeled examples, we enforce a consistency between the main decoder predictions and those of the auxiliary decoders, taking as inputs different perturbed versions of the encoder's output, and consequently, improving the encoder's representations. The proposed method is simple and can easily be extended to use additional training signal, such as image-level labels or pixel-level labels across different domains. We perform an ablation study to tease apart the effectiveness of each component, and conduct extensive experiments to demonstrate that our method achieves state-of-the-art results in several datasets.

arXiv:2003.08981공식 코드 · Papers with Code

Local Implicit Grid Representations for 3D Scenes

Shape priors learned from data are commonly used to reconstruct 3D objects from partial or noisy data. Yet no such shape priors are available for indoor scenes, since typical 3D autoencoders cannot handle their scale, complexity, or diversity. In this paper, we introduce Local Implicit Grid Representations, a new 3D shape representation designed for scalability and generality. The motivating idea is that most 3D surfaces share geometric details at some scale -- i.e., at a scale smaller than an entire object and larger than a small patch. We train an autoencoder to learn an embedding of local crops of 3D shapes at that size. Then, we use the decoder as a component in a shape optimization that solves for a set of latent codes on a regular grid of overlapping crops such that an interpolation of the decoded local shapes matches a partial or noisy observation. We demonstrate the value of this proposed approach for 3D surface reconstruction from sparse point observations, showing significantly better results than alternative approaches.

arXiv:2003.09293

U-Det: A Modified U-Net architecture with bidirectional feature network for lung nodule segmentation

Early diagnosis and analysis of lung cancer involve a precise and efficient lung nodule segmentation in computed tomography (CT) images. However, the anonymous shapes, visual features, and surroundings of the nodule in the CT image pose a challenging problem to the robust segmentation of the lung nodules. This article proposes U-Det, a resource-efficient model architecture, which is an end to end deep learning approach to solve the task at hand. It incorporates a Bi-FPN (bidirectional feature network) between the encoder and decoder. Furthermore, it uses Mish activation function and class weights of masks to enhance segmentation efficiency. The proposed model is extensively trained and evaluated on the publicly available LUNA-16 dataset consisting of 1186 lung nodules. The U-Det architecture outperforms the existing U-Net model with the Dice similarity coefficient (DSC) of 82.82% and achieves results comparable to human experts.

arXiv:2003.09210

DIDFuse: Deep Image Decomposition for Infrared and Visible Image Fusion

Infrared and visible image fusion, a hot topic in the field of image processing, aims at obtaining fused images keeping the advantages of source images. This paper proposes a novel auto-encoder (AE) based fusion network. The core idea is that the encoder decomposes an image into background and detail feature maps with low- and high-frequency information, respectively, and that the decoder recovers the original image. To this end, the loss function makes the background/detail feature maps of source images similar/dissimilar. In the test phase, background and detail feature maps are respectively merged via a fusion module, and the fused image is recovered by the decoder. Qualitative and quantitative results illustrate that our method can generate fusion images containing highlighted targets and abundant detail texture information with strong robustness and meanwhile surpass state-of-the-art (SOTA) approaches.

arXiv:2003.09175

3dDepthNet: Point Cloud Guided Depth Completion Network for Sparse Depth and Single Color Image

In this paper, we propose an end-to-end deep learning network named 3dDepthNet, which produces an accurate dense depth image from a single pair of sparse LiDAR depth and color image for robotics and autonomous driving tasks. Based on the dimensional nature of depth images, our network offers a novel 3D-to-2D coarse-to-fine dual densification design that is both accurate and lightweight. Depth densification is first performed in 3D space via point cloud completion, followed by a specially designed encoder-decoder structure that utilizes the projected dense depth from 3D completion and the original RGB-D images to perform 2D image completion. Experiments on the KITTI dataset show our network achieves state-of-art accuracy while being more efficient. Ablation and generalization tests prove that each module in our network has positive influences on the final results, and furthermore, our network is resilient to even sparser depth.

arXiv:2003.08802

Dynamic Multiscale Graph Neural Networks for 3D Skeleton-Based Human Motion Prediction

We propose novel dynamic multiscale graph neural networks (DMGNN) to predict 3D skeleton-based human motions. The core idea of DMGNN is to use a multiscale graph to comprehensively model the internal relations of a human body for motion feature learning. This multiscale graph is adaptive during training and dynamic across network layers. Based on this graph, we propose a multiscale graph computational unit (MGCU) to extract features at individual scales and fuse features across scales. The entire model is action-category-agnostic and follows an encoder-decoder framework. The encoder consists of a sequence of MGCUs to learn motion features. The decoder uses a proposed graph-based gate recurrent unit to generate future poses. Extensive experiments show that the proposed DMGNN outperforms state-of-the-art methods in both short and long-term predictions on the datasets of Human 3.6M and CMU Mocap. We further investigate the learned multiscale graphs for the interpretability. The codes could be downloaded from https://github.com/limaosen0/DMGNN.

arXiv:2003.08550

Detecting Lane and Road Markings at A Distance with Perspective Transformer Layers

Accurate detection of lane and road markings is a task of great importance for intelligent vehicles. In existing approaches, the detection accuracy often degrades with the increasing distance. This is due to the fact that distant lane and road markings occupy a small number of pixels in the image, and scales of lane and road markings are inconsistent at various distances and perspectives. The Inverse Perspective Mapping (IPM) can be used to eliminate the perspective distortion, but the inherent interpolation can lead to artifacts especially around distant lane and road markings and thus has a negative impact on the accuracy of lane marking detection and segmentation. To solve this problem, we adopt the Encoder-Decoder architecture in Fully Convolutional Networks and leverage the idea of Spatial Transformer Networks to introduce a novel semantic segmentation neural network. This approach decomposes the IPM process into multiple consecutive differentiable homographic transform layers, which are called "Perspective Transformer Layers". Furthermore, the interpolated feature map is refined by subsequent convolutional layers thus reducing the artifacts and improving the accuracy. The effectiveness of the proposed method in lane marking detection is validated on two public datasets: TuSimple and ApolloScape

arXiv:2003.08436공식 코드 · Papers with Code

Collaborative Distillation for Ultra-Resolution Universal Style Transfer

Universal style transfer methods typically leverage rich representations from deep Convolutional Neural Network (CNN) models (e.g., VGG-19) pre-trained on large collections of images. Despite the effectiveness, its application is heavily constrained by the large model size to handle ultra-resolution images given limited memory. In this work, we present a new knowledge distillation method (named Collaborative Distillation) for encoder-decoder based neural style transfer to reduce the convolutional filters. The main idea is underpinned by a finding that the encoder-decoder pairs construct an exclusive collaborative relationship, which is regarded as a new kind of knowledge for style transfer models. Moreover, to overcome the feature size mismatch when applying collaborative distillation, a linear embedding loss is introduced to drive the student network to learn a linear embedding of the teacher's features. Extensive experiments show the effectiveness of our method when applied to different universal style transfer approaches (WCT and AdaIN), even if the model size is reduced by 15.5 times. Especially, on WCT with the compressed models, we achieve ultra-resolution (over 40 megapixels) universal style transfer on a 12GB GPU for the first time. Further experiments on optimization-based stylization scheme show the generality of our algorithm on different stylization paradigms. Our code and trained models are available at https://github.com/mingsun-tse/collaborative-distillation.

arXiv:2003.08407

A Content Transformation Block For Image Style Transfer

Style transfer has recently received a lot of attention, since it allows to study fundamental challenges in image understanding and synthesis. Recent work has significantly improved the representation of color and texture and computational speed and image resolution. The explicit transformation of image content has, however, been mostly neglected: while artistic style affects formal characteristics of an image, such as color, shape or texture, it also deforms, adds or removes content details. This paper explicitly focuses on a content-and style-aware stylization of a content image. Therefore, we introduce a content transformation module between the encoder and decoder. Moreover, we utilize similar content appearing in photographs and style samples to learn how style alters content details and we generalize this to other class details. Additionally, this work presents a novel normalization layer critical for high resolution image synthesis. The robustness and speed of our model enables a video stylization in real-time and high definition. We perform extensive qualitative and quantitative evaluations to demonstrate the validity of our approach.

arXiv:2003.07962

Deliberation Model Based Two-Pass End-to-End Speech Recognition

End-to-end (E2E) models have made rapid progress in automatic speech recognition (ASR) and perform competitively relative to conventional models. To further improve the quality, a two-pass model has been proposed to rescore streamed hypotheses using the non-streaming Listen, Attend and Spell (LAS) model while maintaining a reasonable latency. The model attends to acoustics to rescore hypotheses, as opposed to a class of neural correction models that use only first-pass text hypotheses. In this work, we propose to attend to both acoustics and first-pass hypotheses using a deliberation network. A bidirectional encoder is used to extract context information from first-pass hypotheses. The proposed deliberation model achieves 12% relative WER reduction compared to LAS rescoring in Google Voice Search (VS) tasks, and 23% reduction on a proper noun test set. Compared to a large conventional model, our best model performs 21% relatively better for VS. In terms of computational complexity, the deliberation decoder has a larger size than the LAS decoder, and hence requires more computations in second-pass decoding.

arXiv:2003.08235

CAFENet: Class-Agnostic Few-Shot Edge Detection Network

We tackle a novel few-shot learning challenge, which we call few-shot semantic edge detection, aiming to localize crisp boundaries of novel categories using only a few labeled samples. We also present a Class-Agnostic Few-shot Edge detection Network (CAFENet) based on meta-learning strategy. CAFENet employs a semantic segmentation module in small-scale to compensate for lack of semantic information in edge labels. The predicted segmentation mask is used to generate an attention map to highlight the target object region, and make the decoder module concentrate on that region. We also propose a new regularization method based on multi-split matching. In meta-training, the metric-learning problem with high-dimensional vectors are divided into small subproblems with low-dimensional sub-vectors. Since there is no existing dataset for few-shot semantic edge detection, we construct two new datasets, FSE-1000 and SBD-$5^i$, and evaluate the performance of the proposed CAFENet on them. Extensive simulation results confirm the performance merits of the techniques adopted in CAFENet.

arXiv:2003.08077

Scene Text Recognition via Transformer

Scene text recognition with arbitrary shape is very challenging due to large variations in text shapes, fonts, colors, backgrounds, etc. Most state-of-the-art algorithms rectify the input image into the normalized image, then treat the recognition as a sequence prediction task. The bottleneck of such methods is the rectification, which will cause errors due to distortion perspective. In this paper, we find that the rectification is completely unnecessary. What all we need is the spatial attention. We therefore propose a simple but extremely effective scene text recognition method based on transformer [50]. Different from previous transformer based models [56,34], which just use the decoder of the transformer to decode the convolutional attention, the proposed method use a convolutional feature maps as word embedding input into transformer. In such a way, our method is able to make full use of the powerful attention mechanism of the transformer. Extensive experimental results show that the proposed method significantly outperforms state-of-the-art methods by a very large margin on both regular and irregular text datasets. On one of the most challenging CUTE dataset whose state-of-the-art prediction accuracy is 89.6%, our method achieves 99.3%, which is a pretty surprising result. We will release our source code and believe that our method will be a new benchmark of scene text recognition with arbitrary shapes.

arXiv:2003.07740

Geometric Approaches to Increase the Expressivity of Deep Neural Networks for MR Reconstruction

Recently, deep learning approaches have been extensively investigated to reconstruct images from accelerated magnetic resonance image (MRI) acquisition. Although these approaches provide significant performance gain compared to compressed sensing MRI (CS-MRI), it is not clear how to choose a suitable network architecture to balance the trade-off between network complexity and performance. Recently, it was shown that an encoder-decoder convolutional neural network (CNN) can be interpreted as a piecewise linear basis-like representation, whose specific representation is determined by the ReLU activation patterns for a given input image. Thus, the expressivity or the representation power is determined by the number of piecewise linear regions. As an extension of this geometric understanding, this paper proposes a systematic geometric approach using bootstrapping and subnetwork aggregation using an attention module to increase the expressivity of the underlying neural network. Our method can be implemented in both k-space domain and image domain that can be trained in an end-to-end manner. Experimental results show that the proposed schemes significantly improve reconstruction performance with negligible complexity increases.