Deep learning technology has made great progress in multi-view 3D reconstruction tasks. At present, most mainstream solutions establish the mapping between views and shape of an object by assembling the networks of 2D encoder and 3D decoder as the basic structure while they adopt different approaches to obtain aggregation of features from several views. Among them, the methods using attention-based fusion perform better and more stable than the others, however, they still have an obvious shortcoming -- the strong independence of each view during predicting the weights for merging leads to a lack of adaption of the global state. In this paper, we propose a global-aware attention-based fusion approach that builds the correlation between each branch and the global to provide a comprehensive foundation for weights inference. In order to enhance the ability of the network, we introduce a novel loss function to supervise the shape overall and propose a dynamic two-stage training strategy that can effectively adapt to all reconstructors with attention-based fusion. Experiments on ShapeNet verify that our method outperforms existing SOTA methods while the amount of parameters is far less than the same type of algorithm, Pix2Vox++. Furthermore, we propose a view-reduction method based on maximizing diversity and discuss the cost-performance tradeoff of our model to achieve a better performance when facing heavy input amount and limited computational cost.
In this work, we propose a streaming AV-ASR system based on a hybrid connectionist temporal classification (CTC)/attention neural network architecture. The audio and the visual encoder neural networks are both based on the conformer architecture, which is made streamable using chunk-wise self-attention (CSA) and causal convolution. Streaming recognition with a decoder neural network is realized by using the triggered attention technique, which performs time-synchronous decoding with joint CTC/attention scoring. Additionally, we propose a novel alignment regularization technique that promotes synchronization of the audio and visual encoder, which in turn results in better word error rates (WERs) at all SNR levels for streaming and offline AV-ASR models. The proposed AV-ASR model achieves WERs of 2.0% and 2.6% on the Lip Reading Sentences 3 (LRS3) dataset in an offline and online setup, respectively, which both present state-of-the-art results when no external training data are used.
In this work, we consider the use of a model-based decoder in combination with an unsupervised learning strategy for direction-of-arrival (DoA) estimation. Relying only on unlabeled training data we show in our analysis that we can outperform existing unsupervised machine learning methods and classical methods. The proposed approach consists of introducing a model-based decoder in an autoencoder architecture which leads to a meaningful representation of the statistical model in the latent space of the autoencoder. Our numerical simulations show that the performance of the presented approach is not affected by correlated signals and performs well for both, uncorrelated and correlated, scenarios. This is a result of the fact, that, in the proposed framework, the signal covariance matrix and the DOAs are estimated simultaneously.
This paper proposes a novel technique to obtain better downstream ASR performance from a joint encoder-decoder self-supervised model when trained with speech pooled from two different channels (narrow and wide band). The joint encoder-decoder self-supervised model extends the HuBERT model with a Transformer decoder. HuBERT performs clustering of features and predicts the class of every input frame. In simple pooling, which is our baseline, there is no way to identify the channel information. To incorporate channel information, we have proposed non-overlapping cluster IDs for speech from different channels. Our method gives a relative improvement of ~4% over the joint encoder-decoder self-supervised model built with simple pooling of data, which serves as our baseline.
In the field of medical CT image processing, convolutional neural networks (CNNs) have been the dominant technique.Encoder-decoder CNNs utilise locality for efficiency, but they cannot simulate distant pixel interactions properly.Recent research indicates that self-attention or transformer layers can be stacked to efficiently learn long-range dependencies.By constructing and processing picture patches as embeddings, transformers have been applied to computer vision applications. However, transformer-based architectures lack global semantic information interaction and require a large-scale training dataset, making it challenging to train with small data samples. In order to solve these challenges, we present a hierarchical contextattention transformer network (MHITNet) that combines the multi-scale, transformer, and hierarchical context extraction modules in skip-connections. The multi-scale module captures deeper CT semantic information, enabling transformers to encode feature maps of tokenized picture patches from various CNN stages as input attention sequences more effectively. The hierarchical context attention module augments global data and reweights pixels to capture semantic context.Extensive trials on three datasets show that the proposed MHITNet beats current best practises
Motor brain-computer interface (BCI) development relies critically on neural time series decoding algorithms. Recent advances in deep learning architectures allow for automatic feature selection to approximate higher-order dependencies in data. This article presents the FingerFlex model - a convolutional encoder-decoder architecture adapted for finger movement regression on electrocorticographic (ECoG) brain data. State-of-the-art performance was achieved on a publicly available BCI competition IV dataset 4 with a correlation coefficient between true and predicted trajectories up to 0.74. The presented method provides the opportunity for developing fully-functional high-precision cortical motor brain-computer interfaces.
We present PolyBuilding, a fully end-to-end polygon Transformer for building extraction. PolyBuilding direct predicts vector representation of buildings from remote sensing images. It builds upon an encoder-decoder transformer architecture and simultaneously outputs building bounding boxes and polygons. Given a set of polygon queries, the model learns the relations among them and encodes context information from the image to predict the final set of building polygons with fixed vertex numbers. Corner classification is performed to distinguish the building corners from the sampled points, which can be used to remove redundant vertices along the building walls during inference. A 1-d non-maximum suppression (NMS) is further applied to reduce vertex redundancy near the building corners. With the refinement operations, polygons with regular shapes and low complexity can be effectively obtained. Comprehensive experiments are conducted on the CrowdAI dataset. Quantitative and qualitative results show that our approach outperforms prior polygonal building extraction methods by a large margin. It also achieves a new state-of-the-art in terms of pixel-level coverage, instance-level precision and recall, and geometry-level properties (including contour regularity and polygon complexity).
In recent years, deep convolution neural networks (DCNNs) have achieved great prospects in coronary artery vessel segmentation. However, it is difficult to deploy complicated models in clinical scenarios since high-performance approaches have excessive parameters and high computation costs. To tackle this problem, we propose \textbf{LightVessel}, a Similarity Knowledge Distillation Framework, for lightweight coronary artery vessel segmentation. Primarily, we propose a Feature-wise Similarity Distillation (FSD) module for semantic-shift modeling. Specifically, we calculate the feature similarity between the symmetric layers from the encoder and decoder. Then the similarity is transferred as knowledge from a cumbersome teacher network to a non-trained lightweight student network. Meanwhile, for encouraging the student model to learn more pixel-wise semantic information, we introduce the Adversarial Similarity Distillation (ASD) module. Concretely, the ASD module aims to construct the spatial adversarial correlation between the annotation and prediction from the teacher and student models, respectively. Through the ASD module, the student model obtains fined-grained subtle edge segmented results of the coronary artery vessel. Extensive experiments conducted on Clinical Coronary Artery Vessel Dataset demonstrate that LightVessel outperforms various knowledge distillation counterparts.
Syntactically controlled paraphrase generation has become an emerging research direction in recent years. Most existing approaches require annotated paraphrase pairs for training and are thus costly to extend to new domains. Unsupervised approaches, on the other hand, do not need paraphrase pairs but suffer from relatively poor performance in terms of syntactic control and quality of generated paraphrases. In this paper, we demonstrate that leveraging Abstract Meaning Representations (AMR) can greatly improve the performance of unsupervised syntactically controlled paraphrase generation. Our proposed model, AMR-enhanced Paraphrase Generator (AMRPG), separately encodes the AMR graph and the constituency parse of the input sentence into two disentangled semantic and syntactic embeddings. A decoder is then learned to reconstruct the input sentence from the semantic and syntactic embeddings. Our experiments show that AMRPG generates more accurate syntactically controlled paraphrases, both quantitatively and qualitatively, compared to the existing unsupervised approaches. We also demonstrate that the paraphrases generated by AMRPG can be used for data augmentation to improve the robustness of NLP models.
This paper presents InterMPL, a semi-supervised learning method of end-to-end automatic speech recognition (ASR) that performs pseudo-labeling (PL) with intermediate supervision. Momentum PL (MPL) trains a connectionist temporal classification (CTC)-based model on unlabeled data by continuously generating pseudo-labels on the fly and improving their quality. In contrast to autoregressive formulations, such as the attention-based encoder-decoder and transducer, CTC is well suited for MPL, or PL-based semi-supervised ASR in general, owing to its simple/fast inference algorithm and robustness against generating collapsed labels. However, CTC generally yields inferior performance than the autoregressive models due to the conditional independence assumption, thereby limiting the performance of MPL. We propose to enhance MPL by introducing intermediate loss, inspired by the recent advances in CTC-based modeling. Specifically, we focus on self-conditional and hierarchical conditional CTC, that apply auxiliary CTC losses to intermediate layers such that the conditional independence assumption is explicitly relaxed. We also explore how pseudo-labels should be generated and used as supervision for intermediate losses. Experimental results in different semi-supervised settings demonstrate that the proposed approach outperforms MPL and improves an ASR model by up to a 12.1% absolute performance gain. In addition, our detailed analysis validates the importance of the intermediate loss.
We present BERT-CTC-Transducer (BECTRA), a novel end-to-end automatic speech recognition (E2E-ASR) model formulated by the transducer with a BERT-enhanced encoder. Integrating a large-scale pre-trained language model (LM) into E2E-ASR has been actively studied, aiming to utilize versatile linguistic knowledge for generating accurate text. One crucial factor that makes this integration challenging lies in the vocabulary mismatch; the vocabulary constructed for a pre-trained LM is generally too large for E2E-ASR training and is likely to have a mismatch against a target ASR domain. To overcome such an issue, we propose BECTRA, an extended version of our previous BERT-CTC, that realizes BERT-based E2E-ASR using a vocabulary of interest. BECTRA is a transducer-based model, which adopts BERT-CTC for its encoder and trains an ASR-specific decoder using a vocabulary suitable for a target task. With the combination of the transducer and BERT-CTC, we also propose a novel inference algorithm for taking advantage of both autoregressive and non-autoregressive decoding. Experimental results on several ASR tasks, varying in amounts of data, speaking styles, and languages, demonstrate that BECTRA outperforms BERT-CTC by effectively dealing with the vocabulary mismatch while exploiting BERT knowledge.
In many cutting-edge applications, high-fidelity computational models prove to be too slow for practical use and are therefore replaced by much faster surrogate models. Recently, deep learning techniques have increasingly been utilized to accelerate such predictions. To enable learning on large-dimensional and complex data, specific neural network architectures have been developed, including convolutional and graph neural networks. In this work, we present a novel encoder-decoder geometric deep learning framework called MAgNET, which extends the well-known convolutional neural networks to accommodate arbitrary graph-structured data. MAgNET consists of innovative Multichannel Aggregation (MAg) layers and graph pooling/unpooling layers, forming a graph U-Net architecture that is analogous to convolutional U-Nets. We demonstrate the predictive capabilities of MAgNET in surrogate modeling for non-linear finite element simulations in the mechanics of solids.
Few-shot segmentation (FSS) aims to segment objects of unseen classes given only a few annotated support images. Most existing methods simply stitch query features with independent support prototypes and segment the query image by feeding the mixed features to a decoder. Although significant improvements have been achieved, existing methods are still face class biases due to class variants and background confusion. In this paper, we propose a joint framework that combines more valuable class-aware and class-agnostic alignment guidance to facilitate the segmentation. Specifically, we design a hybrid alignment module which establishes multi-scale query-support correspondences to mine the most relevant class-aware information for each query image from the corresponding support features. In addition, we explore utilizing base-classes knowledge to generate class-agnostic prior mask which makes a distinction between real background and foreground by highlighting all object regions, especially those of unseen classes. By jointly aggregating class-aware and class-agnostic alignment guidance, better segmentation performances are obtained on query images. Extensive experiments on PASCAL-$5^i$ and COCO-$20^i$ datasets demonstrate that our proposed joint framework performs better, especially on the 1-shot setting.
Several SLAM methods benefit from the use of semantic information. Most integrate photometric methods with high-level semantics such as object detection and semantic segmentation. We propose that adding a semantic segmentation decoder in a shared encoder architecture would help the descriptor decoder learn semantic information, improving the feature extractor. This would be a more robust approach than only using high-level semantic information since it would be intrinsically learned in the descriptor and would not depend on the final quality of the semantic prediction. To add this information, we take advantage of multi-task learning methods to improve accuracy and balance the performance of each task. The proposed models are evaluated according to detection and matching metrics on the HPatches dataset. The results show that the Semantic SuperPoint model performs better than the baseline one.
Deep image compression performs better than conventional codecs, such as JPEG, on natural images. However, deep image compression is learning-based and encounters a problem: the compression performance deteriorates significantly for out-of-domain images. In this study, we highlight this problem and address a novel task: universal deep image compression. This task aims to compress images belonging to arbitrary domains, such as natural images, line drawings, and comics. To address this problem, we propose a content-adaptive optimization framework; this framework uses a pre-trained compression model and adapts the model to a target image during compression. Adapters are inserted into the decoder of the model. For each input image, our framework optimizes the latent representation extracted by the encoder and the adapter parameters in terms of rate-distortion. The adapter parameters are additionally transmitted per image. For the experiments, a benchmark dataset containing uncompressed images of four domains (natural images, line drawings, comics, and vector arts) is constructed and the proposed universal deep compression is evaluated. Finally, the proposed model is compared with non-adaptive and existing adaptive compression models. The comparison reveals that the proposed model outperforms these. The code and dataset are publicly available at https://github.com/kktsubota/universal-dic.
Recent approaches to empathetic response generation incorporate emotion causalities to enhance comprehension of both the user's feelings and experiences. However, these approaches suffer from two critical issues. First, they only consider causalities between the user's emotion and the user's experiences, and ignore those between the user's experiences. Second, they neglect interdependence among causalities and reason them independently. To solve the above problems, we expect to reason all plausible causalities interdependently and simultaneously, given the user's emotion, dialogue history, and future dialogue content. Then, we infuse these causalities into response generation for empathetic responses. Specifically, we design a new model, i.e., the Conditional Variational Graph Auto-Encoder (CVGAE), for the causality reasoning, and adopt a multi-source attention mechanism in the decoder for the causality infusion. We name the whole framework as CARE, abbreviated for CAusality Reasoning for Empathetic conversation. Experimental results indicate that our method achieves state-of-the-art performance.
We study the interpolation capabilities of implicit neural representations (INRs) of images. In principle, INRs promise a number of advantages, such as continuous derivatives and arbitrary sampling, being freed from the restrictions of a raster grid. However, empirically, INRs have been observed to poorly interpolate between the pixels of the fit image; in other words, they do not inherently possess a suitable prior for natural images. In this paper, we propose to address and improve INRs' interpolation capabilities by explicitly integrating image prior information into the INR architecture via deep decoder, a specific implementation of the deep image prior (DIP). Our method, which we call TITAN, leverages a residual connection from the input which enables integrating the principles of the grid-based DIP into the grid-free INR. Through super-resolution and computed tomography experiments, we demonstrate that our method significantly improves upon classic INRs, thanks to the induced natural image bias. We also find that by constraining the weights to be sparse, image quality and sharpness are enhanced, increasing the Lipschitz constant.
There is a critical need for efficient and reliable active flow control strategies to reduce drag and noise in aerospace and marine engineering applications. While traditional full-order models based on the Navier-Stokes equations are not feasible, advanced model reduction techniques can be inefficient for active control tasks, especially with strong non-linearity and convection-dominated phenomena. Using convolutional recurrent autoencoder network architectures, deep learning-based reduced-order models have been recently shown to be effective while performing several orders of magnitude faster than full-order simulations. However, these models encounter significant challenges outside the training data, limiting their effectiveness for active control and optimization tasks. In this study, we aim to improve the extrapolation capability by modifying network architecture and integrating coupled space-time physics as an implicit bias. Reduced-order models via deep learning generally employ decoupling in spatial and temporal dimensions, which can introduce modeling and approximation errors. To alleviate these errors, we propose a novel technique for learning coupled spatial-temporal correlation using a 3D convolution network. We assess the proposed technique against a standard encoder-propagator-decoder model and demonstrate a superior extrapolation performance. To demonstrate the effectiveness of 3D convolution network, we consider a benchmark problem of the flow past a circular cylinder at laminar flow conditions and use the spatio-temporal snapshots from the full-order simulations. Our proposed 3D convolution architecture accurately captures the velocity and pressure fields for varying Reynolds numbers. Compared to the standard encoder-propagator-decoder network, the spatio-temporal-based 3D convolution network improves the prediction range of Reynolds numbers outside of the training data.
Unsupervised visual anomaly detection conveys practical significance in many scenarios and is a challenging task due to the unbounded definition of anomalies. Moreover, most previous methods are application-specific, and establishing a unified model for anomalies across application scenarios remains unsolved. This paper proposes a novel hybrid framework termed Siamese Transition Masked Autoencoders(ST-MAE) to handle various visual anomaly detection tasks uniformly via deep feature transition. Concretely, the proposed method first extracts hierarchical semantics features from a pre-trained deep convolutional neural network and then develops a feature decoupling strategy to split the deep features into two disjoint feature patch subsets. Leveraging the decoupled features, the ST-MAE is developed with the Siamese encoders that operate on each subset of feature patches and perform the latent representations transition of two subsets, along with a lightweight decoder that reconstructs the original feature from the transitioned latent representation. Finally, the anomalous attributes can be detected using the semantic deep feature residual. Our deep feature transition scheme yields a nontrivial and semantic self-supervisory task to extract prototypical normal patterns, which allows for learning uniform models that generalize well for different visual anomaly detection tasks. The extensive experiments conducted demonstrate that the proposed ST-MAE method can advance state-of-the-art performance on multiple benchmarks across application scenarios with a superior inference efficiency, which exhibits great potential to be the uniform model for unsupervised visual anomaly detection.
We consider the task of image-captioning using only the CLIP model and additional text data at training time, and no additional captioned images. Our approach relies on the fact that CLIP is trained to make visual and textual embeddings similar. Therefore, we only need to learn how to translate CLIP textual embeddings back into text, and we can learn how to do this by learning a decoder for the frozen CLIP text encoder using only text. We argue that this intuition is "almost correct" because of a gap between the embedding spaces, and propose to rectify this via noise injection during training. We demonstrate the effectiveness of our approach by showing SOTA zero-shot image captioning across four benchmarks, including style transfer. Code, data, and models are available on GitHub.