For 3D hand and body pose estimation task in depth image, a novel anchor-based approach termed Anchor-to-Joint regression network (A2J) with the end-to-end learning ability is proposed. Within A2J, anchor points able to capture global-local spatial context information are densely set on depth image as local regressors for the joints. They contribute to predict the positions of the joints in ensemble way to enhance generalization ability. The proposed 3D articulated pose estimation paradigm is different from the state-of-the-art encoder-decoder based FCN, 3D CNN and point-set based manners. To discover informative anchor points towards certain joint, anchor proposal procedure is also proposed for A2J. Meanwhile 2D CNN (i.e., ResNet-50) is used as backbone network to drive A2J, without using time-consuming 3D convolutional or deconvolutional layers. The experiments on 3 hand datasets and 2 body datasets verify A2J's superiority. Meanwhile, A2J is of high running speed around 100 FPS on single NVIDIA 1080Ti GPU.
We show that existing upsampling operators can be unified using the notion of the index function. This notion is inspired by an observation in the decoding process of deep image matting where indices-guided unpooling can often recover boundary details considerably better than other upsampling operators such as bilinear interpolation. By viewing the indices as a function of the feature map, we introduce the concept of "learning to index", and present a novel index-guided encoder-decoder framework where indices are self-learned adaptively from data and are used to guide the downsampling and upsampling stages, without extra training supervision. At the core of this framework is a new learnable module, termed Index Network (IndexNet), which dynamically generates indices conditioned on the feature map itself. IndexNet can be used as a plug-in applying to almost all off-the-shelf convolutional networks that have coupled downsampling and upsampling stages, giving the networks the ability to dynamically capture variations of local patterns. In particular, we instantiate and investigate five families of IndexNet and demonstrate their effectiveness on four dense prediction tasks, including image denoising, image matting, semantic segmentation, and monocular depth estimation. Code and models have been made available at: https://tinyurl.com/IndexNetV1
Scene text recognition has attracted particular research interest because it is a very challenging problem and has various applications. The most cutting-edge methods are attentional encoder-decoder frameworks that learn the alignment between the input image and output sequences. In particular, the decoder recurrently outputs predictions, using the prediction of the previous step as a guidance for every time step. In this study, we point out that the inappropriate use of previous predictions in existing attention mechanisms restricts the recognition performance and brings instability. To handle this problem, we propose a novel module, namely adaptive embedding gate(AEG). The proposed AEG focuses on introducing high-order character language models to attention mechanism by controlling the information transmission between adjacent characters. AEG is a flexible module and can be easily integrated into the state-of-the-art attentional methods. We evaluate its effectiveness as well as robustness on a number of standard benchmarks, including the IIIT$5$K, SVT, SVT-P, CUTE$80$, and ICDAR datasets. Experimental results demonstrate that AEG can significantly boost recognition performance and bring better robustness.
This paper proposes a novel procedure for training an encoder-decoder based deep neural network which compresses NxM models into a single model enabling us to dynamically choose the number of encoder and decoder layers for decoding. Usually, the output of the last layer of the N-layer encoder is fed to the M-layer decoder, and the output of the last decoder layer is used to compute softmax loss. Instead, our method computes a single loss consisting of NxM losses: the softmax loss for the output of each of the M decoder layers derived using the output of each of the N encoder layers. A single model trained by our method can be used for decoding with an arbitrary fewer number of encoder and decoder layers. In practical scenarios, this (a) enables faster decoding with insignificant losses in translation quality and (b) alleviates the need to train NxM models, thereby saving space. We take a case study of neural machine translation and show the advantage and give a cost-benefit analysis of our approach.
In this paper, we propose to guide the video caption generation with Part-of-Speech (POS) information, based on a gated fusion of multiple representations of input videos. We construct a novel gated fusion network, with one particularly designed cross-gating (CG) block, to effectively encode and fuse different types of representations, e.g., the motion and content features of an input video. One POS sequence generator relies on this fused representation to predict the global syntactic structure, which is thereafter leveraged to guide the video captioning generation and control the syntax of the generated sentence. Specifically, a gating strategy is proposed to dynamically and adaptively incorporate the global syntactic POS information into the decoder for generating each word. Experimental results on two benchmark datasets, namely MSR-VTT and MSVD, demonstrate that the proposed model can well exploit complementary information from multiple representations, resulting in improved performances. Moreover, the generated global POS information can well capture the global syntactic structure of the sentence, and thus be exploited to control the syntactic structure of the description. Such POS information not only boosts the video captioning performance but also improves the diversity of the generated captions. Our code is at: https://github.com/vsislab/Controllable_XGating.
Constructing powerful generative models for natural images is a challenging task. PixelCNN models capture details and local information in images very well but have limited receptive field. Variational autoencoders with a factorial decoder can capture global information easily, but they often fail to reconstruct details faithfully. PixelVAE combines the best features of the two models and constructs a generative model that is able to learn local and global structures. Here we introduce PixelVAE++, a VAE with three types of latent variables and a PixelCNN++ for the decoder. We introduce a novel architecture that reuses a part of the decoder as an encoder. We achieve the state of the art performance on binary data sets such as MNIST and Omniglot and achieve the state of the art performance on CIFAR-10 among latent variable models while keeping the latent variables informative.
Multi-scale context module and single-stage encoder-decoder structure are commonly employed for semantic segmentation. The multi-scale context module refers to the operations to aggregate feature responses from a large spatial extent, while the single-stage encoder-decoder structure encodes the high-level semantic information in the encoder path and recovers the boundary information in the decoder path. In contrast, multi-stage encoder-decoder networks have been widely used in human pose estimation and show superior performance than their single-stage counterpart. However, few efforts have been attempted to bring this effective design to semantic segmentation. In this work, we propose a Semantic Prediction Guidance (SPG) module which learns to re-weight the local features through the guidance from pixel-wise semantic prediction. We find that by carefully re-weighting features across stages, a two-stage encoder-decoder network coupled with our proposed SPG module can significantly outperform its one-stage counterpart with similar parameters and computations. Finally, we report experimental results on the semantic segmentation benchmark Cityscapes, in which our SPGNet attains 81.1% on the test set using only 'fine' annotations.
We propose a novel approach for semantic segmentation that uses an encoder in the reverse direction to decode. Many semantic segmentation networks adopt a feedforward encoder-decoder architecture. Typically, an input is first downsampled by the encoder to extract high-level semantic features and continues to be fed forward through the decoder module to recover low-level spatial clues. Our method works in an alternative direction that lets information flow backward from the last layer of the encoder towards the first. The encoder performs encoding in the forward pass and the same network performs decoding in the backward pass. Therefore, the encoder itself is also the decoder. Compared to conventional encoder-decoder architectures, ours doesn't require additional layers for decoding and further reuses the encoder weights thereby reducing the total number of parameters required for processing. We show by using only the 13 convolutional layers from VGG-16 plus one tiny classification layer, our model significantly outperforms other frequently cited models that are also adapted from VGG-16. On the Cityscapes semantic segmentation benchmark, our model uses 50.0% less parameters than SegNet and achieves an 18.1% higher "IoU class" score; it uses 28.3% less parameters than DeepLab LargeFOV and the achieved "IoU class" score is 3.9% higher; it uses 89.1% fewer parameters than FCN-8s and the achieved "IoU class" score is 3.1% higher. Our code will be publicly available on Github later.
Recent works on relational triple extraction have shown the superiority of jointly extracting entities and relations over the pipelined extraction manner. However, most existing joint models fail to balance the modeling of entity features and the joint decoding strategy, and thus the interactions between the entity level and triple level are not fully investigated. In this work, we first introduce the hierarchical dependency and horizontal commonality between the two levels, and then propose an entity-enhanced dual tagging framework that enables the triple extraction (TE) task to utilize such interactions with self-learned entity features through an auxiliary entity extraction (EE) task, without breaking the joint decoding of relational triples. Specifically, we align the EE and TE tasks in a position-wise manner by formulating them as two sequence labeling problems with identical encoder-decoder structure. Moreover, the two tasks are organized in a carefully designed parameter sharing setting so that the learned entity features could be naturally shared via multi-task learning. Empirical experiments on the NYT benchmark demonstrate the effectiveness of the proposed framework compared to the state-of-the-art methods.
This paper presents a new sequence-to-sequence (seq2seq) pre-training method PoDA (Pre-training of Denoising Autoencoders), which learns representations suitable for text generation tasks. Unlike encoder-only (e.g., BERT) or decoder-only (e.g., OpenAI GPT) pre-training approaches, PoDA jointly pre-trains both the encoder and decoder by denoising the noise-corrupted text, and it also has the advantage of keeping the network architecture unchanged in the subsequent fine-tuning stage. Meanwhile, we design a hybrid model of Transformer and pointer-generator networks as the backbone architecture for PoDA. We conduct experiments on two text generation tasks: abstractive summarization, and grammatical error correction. Results on four datasets show that PoDA can improve model performance over strong baselines without using any task-specific techniques and significantly speed up convergence.
Surface mount technology (SMT) is a process for producing printed circuit boards. Solder paste printer (SPP), package mounter, and solder reflow oven are used for SMT. The board on which the solder paste is deposited from the SPP is monitored by solder paste inspector (SPI). If SPP malfunctions due to the printer defects, the SPP produces defective products, and then abnormal patterns are detected by SPI. In this paper, we propose a convolutional recurrent reconstructive network (CRRN), which decomposes the anomaly patterns generated by the printer defects, from SPI data. CRRN learns only normal data and detects anomaly pattern through reconstruction error. CRRN consists of a spatial encoder (S-Encoder), a spatiotemporal encoder and decoder (ST-Encoder-Decoder), and a spatial decoder (S-Decoder). The ST-Encoder-Decoder consists of multiple convolutional spatiotemporal memories (CSTMs) with ST-Attention mechanism. CSTM is developed to extract spatiotemporal patterns efficiently. Additionally, a spatiotemporal attention (ST-Attention) mechanism is designed to facilitate transmitting information from the ST-Encoder to the ST-Decoder, which can solve the long-term dependency problem. We demonstrate the proposed CRRN outperforms the other conventional models in anomaly detection. Moreover, we show the discriminative power of the anomaly map decomposed by the proposed CRRN through the printer defect classification.
This paper focuses on the task of sentiment transfer on non-parallel text, which modifies sentiment attributes (e.g., positive or negative) of sentences while preserving their attribute-independent content. Due to the limited capability of RNNbased encoder-decoder structure to capture deep and long-range dependencies among words, previous works can hardly generate satisfactory sentences from scratch. When humans convert the sentiment attribute of a sentence, a simple but effective approach is to only replace the original sentimental tokens in the sentence with target sentimental expressions, instead of building a new sentence from scratch. Such a process is very similar to the task of Text Infilling or Cloze, which could be handled by a deep bidirectional Masked Language Model (e.g. BERT). So we propose a two step approach "Mask and Infill". In the mask step, we separate style from content by masking the positions of sentimental tokens. In the infill step, we retrofit MLM to Attribute Conditional MLM, to infill the masked positions by predicting words or phrases conditioned on the context1 and target sentiment. We evaluate our model on two review datasets with quantitative, qualitative, and human evaluations. Experimental results demonstrate that our models improve state-of-the-art performance.
Online health communities such as the online breast cancer forum enable patients (i.e., users) to interact and help each other within various subforums, which are subsections of the main forum devoted to specific health topics. The changing nature of the users' activities in different subforums can be strong indicators of their health status changes. This additional information could allow health-care organizations to respond promptly and provide additional help for the patient. However, modeling complex transitions of an individual user's activities among different subforums over time and learning how these correspond to his/her health stage are extremely challenging. In this paper, we first formulate the transition of user activities as a dynamic graph with multi-attributed nodes, then formalize the health stage inference task as a dynamic graph-to-sequence learning problem, and hence propose a novel dynamic graph-to-sequence neural networks architecture (DynGraph2Seq) to address all the challenges. Our proposed DynGraph2Seq model consists of a novel dynamic graph encoder and an interpretable sequence decoder that learn the mapping between a sequence of time-evolving user activity graphs and a sequence of target health stages. We go on to propose dynamic graph hierarchical attention mechanisms to facilitate the necessary multi-level interpretability. A comprehensive experimental analysis of its use for a health stage prediction task demonstrates both the effectiveness and the interpretability of the proposed models.
Bidirectional Encoder Representations from Transformers (BERT) represents the latest incarnation of pretrained language models which have recently advanced a wide range of natural language processing tasks. In this paper, we showcase how BERT can be usefully applied in text summarization and propose a general framework for both extractive and abstractive models. We introduce a novel document-level encoder based on BERT which is able to express the semantics of a document and obtain representations for its sentences. Our extractive model is built on top of this encoder by stacking several inter-sentence Transformer layers. For abstractive summarization, we propose a new fine-tuning schedule which adopts different optimizers for the encoder and the decoder as a means of alleviating the mismatch between the two (the former is pretrained while the latter is not). We also demonstrate that a two-staged fine-tuning approach can further boost the quality of the generated summaries. Experiments on three datasets show that our model achieves state-of-the-art results across the board in both extractive and abstractive settings. Our code is available at https://github.com/nlpyang/PreSumm
Left ventricle segmentation and morphological assessment are essential for improving diagnosis and our understanding of cardiomyopathy, which in turn is imperative for reducing risk of myocardial infarctions in patients. Convolutional neural network (CNN) based methods for cardiac magnetic resonance (CMR) image segmentation rely on supervision with pixel-level annotations, and may not generalize well to images from a different domain. These methods are typically sensitive to variations in imaging protocols and data acquisition. Since annotating multi-sequence CMR images is tedious and subject to inter- and intra-observer variations, developing methods that can automatically adapt from one domain to the target domain is of great interest. In this paper, we propose an approach for domain adaptation in multi-sequence CMR segmentation task using transfer learning that combines multi-source image information. We first train an encoder-decoder CNN on T2-weighted and balanced-Steady State Free Precession (bSSFP) MR images with pixel-level annotation and fine-tune the same network with a limited number of Late Gadolinium Enhanced-MR (LGE-MR) subjects, to adapt the domain features. The domain-adapted network was trained with just four LGE-MR training samples and obtained an average Dice score of $\sim$85.0\% on the test set comprises of 40 LGE-MR subjects. The proposed method significantly outperformed a network without adaptation trained from scratch on the same set of LGE-MR training data.
Dialogue state tracking (DST) is an essential component in task-oriented dialogue systems, which estimates user goals at every dialogue turn. However, most previous approaches usually suffer from the following problems. Many discriminative models, especially end-to-end (E2E) models, are difficult to extract unknown values that are not in the candidate ontology; previous generative models, which can extract unknown values from utterances, degrade the performance due to ignoring the semantic information of pre-defined ontology. Besides, previous generative models usually need a hand-crafted list to normalize the generated values. How to integrate the semantic information of pre-defined ontology and dialogue text (heterogeneous texts) to generate unknown values and improve performance becomes a severe challenge. In this paper, we propose a Copy-Enhanced Heterogeneous Information Learning model with multiple encoder-decoder for DST (CEDST), which can effectively generate all possible values including unknown values by copying values from heterogeneous texts. Meanwhile, CEDST can effectively decompose the large state space into several small state spaces through multi-encoder, and employ multi-decoder to make full use of the reduced spaces to generate values. Multi-encoder-decoder architecture can significantly improve performance. Experiments show that CEDST can achieve state-of-the-art results on two datasets and our constructed datasets with many unknown values.
In this paper, we propose a method that estimates a gait index for a sequence of skeletons. Our system is a stack of an encoder and a decoder that are formed by Long Short-Term Memories (LSTMs). In the encoding stage, the characteristics of an input are automatically determined and are compressed into a latent space. The decoding stage then attempts to reconstruct the input according to such intermediate representation. The reconstruction error is thus considered as a weak gait index. By combining such weak indices over a long-time movement, our system can provide a good estimation for the gait index. Our experiments on a large dataset (nearly one hundred thousand skeletons) showed that the index given by the proposed method outperformed some recent works on gait analysis.
By benefiting from perceptual losses, recent studies have improved significantly the performance of the super-resolution task, where a high-resolution image is resolved from its low-resolution counterpart. Although such objective functions generate near-photorealistic results, their capability is limited, since they estimate the reconstruction error for an entire image in the same way, without considering any semantic information. In this paper, we propose a novel method to benefit from perceptual loss in a more objective way. We optimize a deep network-based decoder with a targeted objective function that penalizes images at different semantic levels using the corresponding terms. In particular, the proposed method leverages our proposed OBB (Object, Background and Boundary) labels, generated from segmentation labels, to estimate a suitable perceptual loss for boundaries, while considering texture similarity for backgrounds. We show that our proposed approach results in more realistic textures and sharper edges, and outperforms other state-of-the-art algorithms in terms of both qualitative results on standard benchmarks and results of extensive user studies.
Semi-implicit graph variational auto-encoder (SIG-VAE) is proposed to expand the flexibility of variational graph auto-encoders (VGAE) to model graph data. SIG-VAE employs a hierarchical variational framework to enable neighboring node sharing for better generative modeling of graph dependency structure, together with a Bernoulli-Poisson link decoder. Not only does this hierarchical construction provide a more flexible generative graph model to better capture real-world graph properties, but also does SIG-VAE naturally lead to semi-implicit hierarchical variational inference that allows faithful modeling of implicit posteriors of given graph data, which may exhibit heavy tails, multiple modes, skewness, and rich dependency structures. Compared to VGAE, the derived graph latent representations by SIG-VAE are more interpretable, due to more expressive generative model and more faithful inference enabled by the flexible semi-implicit construction. Extensive experiments with a variety of graph data show that SIG-VAE significantly outperforms state-of-the-art methods on several different graph analytic tasks.
Steady progress has been made in abstractive summarization with attention-based sequence-to-sequence learning models. In this paper, we propose a new decoder where the output summary is generated by conditioning on both the input text and the latent topics of the document. The latent topics, identified by a topic model such as LDA, reveals more global semantic information that can be used to bias the decoder to generate words. In particular, they enable the decoder to have access to additional word co-occurrence statistics captured at document corpus level. We empirically validate the advantage of the proposed approach on both the CNN/Daily Mail and the WikiHow datasets. Concretely, we attain strongly improved ROUGE scores when compared to state-of-the-art models.