PyTorchKR Papers

최신 논문

총 9,642편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2201.07451

TransFuse: A Unified Transformer-based Image Fusion Framework using Self-supervised Learning

Image fusion is a technique to integrate information from multiple source images with complementary information to improve the richness of a single image. Due to insufficient task-specific training data and corresponding ground truth, most existing end-to-end image fusion methods easily fall into overfitting or tedious parameter optimization processes. Two-stage methods avoid the need of large amount of task-specific training data by training encoder-decoder network on large natural image datasets and utilizing the extracted features for fusion, but the domain gap between natural images and different fusion tasks results in limited performance. In this study, we design a novel encoder-decoder based image fusion framework and propose a destruction-reconstruction based self-supervised training scheme to encourage the network to learn task-specific features. Specifically, we propose three destruction-reconstruction self-supervised auxiliary tasks for multi-modal image fusion, multi-exposure image fusion and multi-focus image fusion based on pixel intensity non-linear transformation, brightness transformation and noise transformation, respectively. In order to encourage different fusion tasks to promote each other and increase the generalizability of the trained network, we integrate the three self-supervised auxiliary tasks by randomly choosing one of them to destroy a natural image in model training. In addition, we design a new encoder that combines CNN and Transformer for feature extraction, so that the trained model can exploit both local and global information. Extensive experiments on multi-modal image fusion, multi-exposure image fusion and multi-focus image fusion tasks demonstrate that our proposed method achieves the state-of-the-art performance in both subjective and objective evaluations. The code will be publicly available soon.

arXiv:2201.07436공식 코드 · Papers with Code

Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth

Depth estimation from a single image is an important task that can be applied to various fields in computer vision, and has grown rapidly with the development of convolutional neural networks. In this paper, we propose a novel structure and training strategy for monocular depth estimation to further improve the prediction accuracy of the network. We deploy a hierarchical transformer encoder to capture and convey the global context, and design a lightweight yet powerful decoder to generate an estimated depth map while considering local connectivity. By constructing connected paths between multi-scale local features and the global decoding stream with our proposed selective feature fusion module, the network can integrate both representations and recover fine details. In addition, the proposed decoder shows better performance than the previously proposed decoders, with considerably less computational complexity. Furthermore, we improve the depth-specific augmentation method by utilizing an important observation in depth estimation to enhance the model. Our network achieves state-of-the-art performance over the challenging depth dataset NYU Depth V2. Extensive experiments have been conducted to validate and show the effectiveness of the proposed approach. Finally, our model shows better generalisation ability and robustness than other comparative models.

arXiv:2201.07006

Time Series Generation with Masked Autoencoder

This paper shows that masked autoencoder with extrapolator (ExtraMAE) is a scalable self-supervised model for time series generation. ExtraMAE randomly masks some patches of the original time series and learns temporal dynamics by recovering the masked patches. Our approach has two core designs. First, ExtraMAE is self-supervised. Supervision allows ExtraMAE to effectively and efficiently capture the temporal dynamics of the original time series. Second, ExtraMAE proposes an extrapolator to disentangle two jobs of the decoder: recovering latent representations and mapping them back into the feature space. These unique designs enable ExtraMAE to consistently and significantly outperform state-of-the-art (SoTA) benchmarks in time series generation. The lightweight architecture also makes ExtraMAE fast and scalable. ExtraMAE shows outstanding behavior in various downstream tasks such as time series classification, prediction, and imputation. As a self-supervised generative model, ExtraMAE allows explicit management of the synthetic data. We hope this paper will usher in a new era of time series generation with self-supervised models.

arXiv:2201.06469

Handling Compounding in Mobile Keyboard Input

This paper proposes a framework to improve the typing experience of mobile users in morphologically rich languages. Smartphone keyboards typically support features such as input decoding, corrections and predictions that all rely on language models. For latency reasons, these operations happen on device, so the models are of limited size and cannot easily cover all the words needed by users for their daily tasks, especially in morphologically rich languages. In particular, the compounding nature of Germanic languages makes their vocabulary virtually infinite. Similarly, heavily inflecting and agglutinative languages (e.g. Slavic, Turkic or Finno-Ugric languages) tend to have much larger vocabularies than morphologically simpler languages, such as English or Mandarin. We propose to model such languages with automatically selected subword units annotated with what we call binding types, allowing the decoder to know when to bind subword units into words. We show that this method brings around 20% word error rate reduction in a variety of compounding languages. This is more than twice the improvement we previously obtained with a more basic approach, also described in the paper.

arXiv:2201.06206공식 코드 · Papers with Code

SQUIRE: A Sequence-to-sequence Framework for Multi-hop Knowledge Graph Reasoning

Multi-hop knowledge graph (KG) reasoning has been widely studied in recent years to provide interpretable predictions on missing links with evidential paths. Most previous works use reinforcement learning (RL) based methods that learn to navigate the path towards the target entity. However, these methods suffer from slow and poor convergence, and they may fail to infer a certain path when there is a missing edge along the path. Here we present SQUIRE, the first Sequence-to-sequence based multi-hop reasoning framework, which utilizes an encoder-decoder Transformer structure to translate the query to a path. Our framework brings about two benefits: (1) It can learn and predict in an end-to-end fashion, which gives better and faster convergence; (2) Our Transformer model does not rely on existing edges to generate the path, and has the flexibility to complete missing edges along the path, especially in sparse KGs. Experiments on standard and sparse KGs show that our approach yields significant improvement over prior methods, while converging 4x-7x faster.

arXiv:2201.05692

Model Stability with Continuous Data Updates

In this paper, we study the "stability" of machine learning (ML) models within the context of larger, complex NLP systems with continuous training data updates. For this study, we propose a methodology for the assessment of model stability (which we refer to as jitter under various experimental conditions. We find that model design choices, including network architecture and input representation, have a critical impact on stability through experiments on four text classification tasks and two sequence labeling tasks. In classification tasks, non-RNN-based models are observed to be more stable than RNN-based ones, while the encoder-decoder model is less stable in sequence labeling tasks. Moreover, input representations based on pre-trained fastText embeddings contribute to more stability than other choices. We also show that two learning strategies -- ensemble models and incremental training -- have a significant influence on stability. We recommend ML model designers account for trade-offs in accuracy and jitter when making modeling choices.

arXiv:2201.05634

Imputing Missing Observations with Time Sliced Synthetic Minority Oversampling Technique

We present a simple yet novel time series imputation technique with the goal of constructing an irregular time series that is uniform across every sample in a data set. Specifically, we fix a grid defined by the midpoints of non-overlapping bins (dubbed "slices") of observation times and ensure that each sample has values for all of the features at that given time. This allows one to both impute fully missing observations to allow uniform time series classification across the entire data and, in special cases, to impute individually missing features. To do so, we slightly generalize the well-known class imbalance algorithm SMOTE \cite{smote} to allow component wise nearest neighbor interpolation that preserves correlations when there are no missing features. We visualize the method in the simplified setting of 2-dimensional uncoupled harmonic oscillators. Next, we use tSMOTE to train an Encoder/Decoder long-short term memory (LSTM) model with Logistic Regression for predicting and classifying distinct trajectories of different 2D oscillators. After illustrating the the utility of tSMOTE in this context, we use the same architecture to train a clinical model for COVID-19 disease severity on an imputed data set. Our experiments show an improvement over standard mean and median imputation techniques by allowing a wider class of patient trajectories to be recognized by the model, as well as improvement over aggregated classification models.

arXiv:2201.05769

Mixed Diagnostics for Longitudinal Properties of Electron Bunches in a Free-Electron Laser

Longitudinal properties of electron bunches are critical for the performance of a wide range of scientific facilities. In a free-electron laser, for example, the existing diagnostics only provide very limited longitudinal information of the electron bunch during online tuning and optimization. We leverage the power of artificial intelligence to build a neural network model using experimental data, in order to bring the destructive longitudinal phase space (LPS) diagnostics online virtually and improve the existing current profile online diagnostics which uses a coherent transition radiation (CTR) spectrometer. The model can also serve as a digital twin of the real machine on which algorithms can be tested efficiently and effectively. We demonstrate at the FLASH facility that the encoder-decoder model with more than one decoder can make highly accurate predictions of megapixel LPS images and coherent transition radiation spectra concurrently for electron bunches in a bunch train with broad ranges of LPS shapes and peak currents, which are obtained by scanning all the major control knobs for LPS manipulation. Furthermore, we propose a way to significantly improve the CTR spectrometer online measurement by combining the predicted and measured spectra. Our work showcases how to combine virtual and real diagnostics in order to provide heterogeneous and reliable mixed diagnostics for scientific facilities.

arXiv:2201.06889

Boosting Robustness of Image Matting with Context Assembling and Strong Data Augmentation

Deep image matting methods have achieved increasingly better results on benchmarks (e.g., Composition-1k/alphamatting.com). However, the robustness, including robustness to trimaps and generalization to images from different domains, is still under-explored. Although some works propose to either refine the trimaps or adapt the algorithms to real-world images via extra data augmentation, none of them has taken both into consideration, not to mention the significant performance deterioration on benchmarks while using those data augmentation. To fill this gap, we propose an image matting method which achieves higher robustness (RMat) via multilevel context assembling and strong data augmentation targeting matting. Specifically, we first build a strong matting framework by modeling ample global information with transformer blocks in the encoder, and focusing on details in combination with convolution layers as well as a low-level feature assembling attention block in the decoder. Then, based on this strong baseline, we analyze current data augmentation and explore simple but effective strong data augmentation to boost the baseline model and contribute a more generalizable matting method. Compared with previous methods, the proposed method not only achieves state-of-the-art results on the Composition-1k benchmark (11% improvement on SAD and 27% improvement on Grad) with smaller model size, but also shows more robust generalization results on other benchmarks, on real-world images, and also on varying coarse-to-fine trimaps with our extensive experiments.

arXiv:2201.06750

DDU-Net: Dual-Decoder-U-Net for Road Extraction Using High-Resolution Remote Sensing Images

Extracting roads from high-resolution remote sensing images (HRSIs) is vital in a wide variety of applications, such as autonomous driving, path planning, and road navigation. Due to the long and thin shape as well as the shades induced by vegetation and buildings, small-sized roads are more difficult to discern. In order to improve the reliability and accuracy of small-sized road extraction when roads of multiple sizes coexist in an HRSI, an enhanced deep neural network model termed Dual-Decoder-U-Net (DDU-Net) is proposed in this paper. Motivated by the U-Net model, a small decoder is added to form a dual-decoder structure for more detailed features. In addition, we introduce the dilated convolution attention module (DCAM) between the encoder and decoders to increase the receptive field as well as to distill multi-scale features through cascading dilated convolution and global average pooling. The convolutional block attention module (CBAM) is also embedded in the parallel dilated convolution and pooling branches to capture more attention-aware features. Extensive experiments are conducted on the Massachusetts Roads dataset with experimental results showing that the proposed model outperforms the state-of-the-art DenseUNet, DeepLabv3+ and D-LinkNet by 6.5%, 3.3%, and 2.1% in the mean Intersection over Union (mIoU), and by 4%, 4.8%, and 3.1% in the F1 score, respectively. Both ablation and heatmap analyses are presented to validate the effectiveness of the proposed model.

arXiv:2201.06390공식 코드 · Papers with Code

SwinUNet3D -- A Hierarchical Architecture for Deep Traffic Prediction using Shifted Window Transformers

Traffic forecasting is an important element of mobility management, an important key that drives the logistics industry. Over the years, lots of work have been done in Traffic forecasting using time series as well as spatiotemporal dynamic forecasting. In this paper, we explore the use of vision transformer in a UNet setting. We completely remove all convolution-based building blocks in UNet, while using 3D shifted window transformer in both encoder and decoder branches. In addition, we experiment with the use of feature mixing just before patch encoding to control the inter-relationship of the feature while avoiding contraction of the depth dimension of our spatiotemporal input. The proposed network is tested on the data provided by Traffic Map Movie Forecasting Challenge 2021(Traffic4cast2021), held in the competition track of Neural Information Processing Systems (NeurIPS). Traffic4cast2021 task is to predict an hour (6 frames) of traffic conditions (volume and average speed)from one hour of given traffic state (12 frames averaged in 5 minutes time span). Source code is available online at https://github.com/bojesomo/Traffic4Cast2021-SwinUNet3D.

arXiv:2201.05963

A Residual Encoder-Decoder Network for Segmentation of Retinal Image-Based Exudates in Diabetic Retinopathy Screening

Diabetic retinopathy refers to the pathology of the retina induced by diabetes and is one of the leading causes of preventable blindness in the world. Early detection of diabetic retinopathy is critical to avoid vision problem through continuous screening and treatment. In traditional clinical practice, the involved lesions are manually detected using photographs of the fundus. However, this task is cumbersome and time-consuming and requires intense effort due to the small size of lesion and low contrast of the images. Thus, computer-assisted diagnosis of diabetic retinopathy based on the detection of red lesions is actively being explored recently. In this paper, we present a convolutional neural network with residual skip connection for the segmentation of exudates in retinal images. To improve the performance of network architecture, a suitable image augmentation technique is used. The proposed network can robustly segment exudates with high accuracy, which makes it suitable for diabetic retinopathy screening. Comparative performance analysis of three benchmark databases: HEI-MED, E-ophtha, and DiaretDB1 is presented. It is shown that the proposed method achieves accuracy (0.98, 0.99, 0.98) and sensitivity (0.97, 0.92, and 0.95) on E-ophtha, HEI-MED, and DiaReTDB1, respectively.

arXiv:2201.05920

ViTBIS: Vision Transformer for Biomedical Image Segmentation

In this paper, we propose a novel network named Vision Transformer for Biomedical Image Segmentation (ViTBIS). Our network splits the input feature maps into three parts with $1\times 1$, $3\times 3$ and $5\times 5$ convolutions in both encoder and decoder. Concat operator is used to merge the features before being fed to three consecutive transformer blocks with attention mechanism embedded inside it. Skip connections are used to connect encoder and decoder transformer blocks. Similarly, transformer blocks and multi scale architecture is used in decoder before being linearly projected to produce the output segmentation map. We test the performance of our network using Synapse multi-organ segmentation dataset, Automated cardiac diagnosis challenge dataset, Brain tumour MRI segmentation dataset and Spleen CT segmentation dataset. Without bells and whistles, our network outperforms most of the previous state of the art CNN and transformer based models using Dice score and the Hausdorff distance as the evaluation metrics.

arXiv:2201.05905

SS-3DCapsNet: Self-supervised 3D Capsule Networks for Medical Segmentation on Less Labeled Data

Capsule network is a recent new deep network architecture that has been applied successfully for medical image segmentation tasks. This work extends capsule networks for volumetric medical image segmentation with self-supervised learning. To improve on the problem of weight initialization compared to previous capsule networks, we leverage self-supervised learning for capsule networks pre-training, where our pretext-task is optimized by self-reconstruction. Our capsule network, SS-3DCapsNet, has a UNet-based architecture with a 3D Capsule encoder and 3D CNNs decoder. Our experiments on multiple datasets including iSeg-2017, Hippocampus, and Cardiac demonstrate that our 3D capsule network with self-supervised pre-training considerably outperforms previous capsule networks and 3D-UNets.

arXiv:2201.05834공식 코드 · Papers with Code

Tailor Versatile Multi-modal Learning for Multi-label Emotion Recognition

Multi-modal Multi-label Emotion Recognition (MMER) aims to identify various human emotions from heterogeneous visual, audio and text modalities. Previous methods mainly focus on projecting multiple modalities into a common latent space and learning an identical representation for all labels, which neglects the diversity of each modality and fails to capture richer semantic information for each label from different perspectives. Besides, associated relationships of modalities and labels have not been fully exploited. In this paper, we propose versaTile multi-modAl learning for multI-labeL emOtion Recognition (TAILOR), aiming to refine multi-modal representations and enhance discriminative capacity of each label. Specifically, we design an adversarial multi-modal refinement module to sufficiently explore the commonality among different modalities and strengthen the diversity of each modality. To further exploit label-modal dependence, we devise a BERT-like cross-modal encoder to gradually fuse private and common modality representations in a granularity descent way, as well as a label-guided decoder to adaptively generate a tailored representation for each label with the guidance of label semantics. In addition, we conduct experiments on the benchmark MMER dataset CMU-MOSEI in both aligned and unaligned settings, which demonstrate the superiority of TAILOR over the state-of-the-arts. Code is available at https://github.com/kniter1/TAILOR.

arXiv:2201.06857

RePre: Improving Self-Supervised Vision Transformer with Reconstructive Pre-training

Recently, self-supervised vision transformers have attracted unprecedented attention for their impressive representation learning ability. However, the dominant method, contrastive learning, mainly relies on an instance discrimination pretext task, which learns a global understanding of the image. This paper incorporates local feature learning into self-supervised vision transformers via Reconstructive Pre-training (RePre). Our RePre extends contrastive frameworks by adding a branch for reconstructing raw image pixels in parallel with the existing contrastive objective. RePre is equipped with a lightweight convolution-based decoder that fuses the multi-hierarchy features from the transformer encoder. The multi-hierarchy features provide rich supervisions from low to high semantic information, which are crucial for our RePre. Our RePre brings decent improvements on various contrastive frameworks with different vision transformer architectures. Transfer performance in downstream tasks outperforms supervised pre-training and state-of-the-art (SOTA) self-supervised counterparts.

arXiv:2201.05596공식 코드 · Papers with Code

DeepSpeed-MoE: Advancing Mixture-of-Experts Inference and Training to Power Next-Generation AI Scale

As the training of giant dense models hits the boundary on the availability and capability of the hardware resources today, Mixture-of-Experts (MoE) models become one of the most promising model architectures due to their significant training cost reduction compared to a quality-equivalent dense model. Its training cost saving is demonstrated from encoder-decoder models (prior works) to a 5x saving for auto-aggressive language models (this work along with parallel explorations). However, due to the much larger model size and unique architecture, how to provide fast MoE model inference remains challenging and unsolved, limiting its practical usage. To tackle this, we present DeepSpeed-MoE, an end-to-end MoE training and inference solution as part of the DeepSpeed library, including novel MoE architecture designs and model compression techniques that reduce MoE model size by up to 3.7x, and a highly optimized inference system that provides 7.3x better latency and cost compared to existing MoE inference solutions. DeepSpeed-MoE offers an unprecedented scale and efficiency to serve massive MoE models with up to 4.5x faster and 9x cheaper inference compared to quality-equivalent dense models. We hope our innovations and systems help open a promising path to new directions in the large model landscape, a shift from dense to sparse MoE models, where training and deploying higher-quality models with fewer resources becomes more widely possible.

arXiv:2201.04066공식 코드 · Papers with Code

VGAER: Graph Neural Network Reconstruction based Community Detection

Community detection is a fundamental and important issue in network science, but there are only a few community detection algorithms based on graph neural networks, among which unsupervised algorithms are almost blank. By fusing the high-order modularity information with network features, this paper proposes a Variational Graph AutoEncoder Reconstruction based community detection VGAER for the first time, and gives its non-probabilistic version. They do not need any prior information. We have carefully designed corresponding input features, decoder, and downstream tasks based on the community detection task and these designs are concise, natural, and perform well (NMI values under our design are improved by 59.1% - 565.9%). Based on a series of experiments with wide range of datasets and advanced methods, VGAER has achieved superior performance and shows strong competitiveness and potential with a simpler design. Finally, we report the results of algorithm convergence analysis and t-SNE visualization, which clearly depicted the stable performance and powerful network modularity ability of VGAER. Our codes are available at https://github.com/qcydm/VGAER.

arXiv:2201.04026

Uni-EDEN: Universal Encoder-Decoder Network by Multi-Granular Vision-Language Pre-training

Vision-language pre-training has been an emerging and fast-developing research topic, which transfers multi-modal knowledge from rich-resource pre-training task to limited-resource downstream tasks. Unlike existing works that predominantly learn a single generic encoder, we present a pre-trainable Universal Encoder-DEcoder Network (Uni-EDEN) to facilitate both vision-language perception (e.g., visual question answering) and generation (e.g., image captioning). Uni-EDEN is a two-stream Transformer based structure, consisting of three modules: object and sentence encoders that separately learns the representations of each modality, and sentence decoder that enables both multi-modal reasoning and sentence generation via inter-modal interaction. Considering that the linguistic representations of each image can span different granularities in this hierarchy including, from simple to comprehensive, individual label, a phrase, and a natural sentence, we pre-train Uni-EDEN through multi-granular vision-language proxy tasks: Masked Object Classification (MOC), Masked Region Phrase Generation (MRPG), Image-Sentence Matching (ISM), and Masked Sentence Generation (MSG). In this way, Uni-EDEN is endowed with the power of both multi-modal representation extraction and language modeling. Extensive experiments demonstrate the compelling generalizability of Uni-EDEN by fine-tuning it to four vision-language perception and generation downstream tasks.

arXiv:2201.04019

Pyramid Fusion Transformer for Semantic Segmentation

The recently proposed MaskFormer gives a refreshed perspective on the task of semantic segmentation: it shifts from the popular pixel-level classification paradigm to a mask-level classification method. In essence, it generates paired probabilities and masks corresponding to category segments and combines them during inference for the segmentation maps. In our study, we find that per-mask classification decoder on top of a single-scale feature is not effective enough to extract reliable probability or mask. To mine for rich semantic information across the feature pyramid, we propose a transformer-based Pyramid Fusion Transformer (PFT) for per-mask approach semantic segmentation with multi-scale features. The proposed transformer decoder performs cross-attention between the learnable queries and each spatial feature from the feature pyramid in parallel and uses cross-scale inter-query attention to exchange complimentary information. We achieve competitive performance on three widely used semantic segmentation datasets. In particular, on ADE20K validation set, our result with Swin-B backbone surpasses that of MaskFormer's with a much larger Swin-L backbone in both single-scale and multi-scale inference, achieving 54.1 mIoU and 55.7 mIoU respectively. Using a Swin-L backbone, we achieve single-scale 56.1 mIoU and multi-scale 57.4 mIoU, obtaining state-of-the-art performance on the dataset. Extensive experiments on three widely used semantic segmentation datasets verify the effectiveness of our proposed method.