PyTorchKR Papers

최신 논문

총 9,339편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2210.15407

Model Order Selection with Variational Autoencoding

Classical methods for model order selection often fail in scenarios with low SNR or few snapshots. Deep learning-based methods are promising alternatives for such challenging situations as they compensate lack of information in the available observations with training on large datasets. This manuscript proposes an approach that uses a variational autoencoder (VAE) for model order selection. The idea is to learn a parameterized conditional covariance matrix at the VAE decoder that approximates the true signal covariance matrix. The method is unsupervised and only requires a small representative dataset for calibration after training the VAE. Numerical simulations show that the proposed method outperforms classical methods and even reaches or beats a supervised approach depending on the considered snapshots.

arXiv:2210.15135

Training Autoregressive Speech Recognition Models with Limited in-domain Supervision

Advances in self-supervised learning have significantly reduced the amount of transcribed audio required for training. However, the majority of work in this area is focused on read speech. We explore limited supervision in the domain of conversational speech. While we assume the amount of in-domain data is limited, we augment the model with open source read speech data. The XLS-R model has been shown to perform well with limited adaptation data and serves as a strong baseline. We use untranscribed data for self-supervised learning and semi-supervised training in an autoregressive encoder-decoder model. We demonstrate that by using the XLS-R model for pseudotranscription, a much smaller autoregressive model can outperform a finetuned XLS-R model when transcribed in-domain data is limited, reducing WER by as much as 8% absolute.

arXiv:2210.15571

Full-scale Deeply Supervised Attention Network for Segmenting COVID-19 Lesions

Automated delineation of COVID-19 lesions from lung CT scans aids the diagnosis and prognosis for patients. The asymmetric shapes and positioning of the infected regions make the task extremely difficult. Capturing information at multiple scales will assist in deciphering features, at global and local levels, to encompass lesions of variable size and texture. We introduce the Full-scale Deeply Supervised Attention Network (FuDSA-Net), for efficient segmentation of corona-infected lung areas in CT images. The model considers activation responses from all levels of the encoding path, encompassing multi-scalar features acquired at different levels of the network. This helps segment target regions (lesions) of varying shape, size and contrast. Incorporation of the entire gamut of multi-scalar characteristics into the novel attention mechanism helps prioritize the selection of activation responses and locations containing useful information. Determining robust and discriminatory features along the decoder path is facilitated with deep supervision. Connections in the decoder arm are remodeled to handle the issue of vanishing gradient. As observed from the experimental results, FuDSA-Net surpasses other state-of-the-art architectures; especially, when it comes to characterizing complicated geometries of the lesions.

arXiv:2210.15392공식 코드 · Papers with Code

LeNo: Adversarial Robust Salient Object Detection Networks with Learnable Noise

Pixel-wise prediction with deep neural network has become an effective paradigm for salient object detection (SOD) and achieved remarkable performance. However, very few SOD models are robust against adversarial attacks which are visually imperceptible for human visual attention. The previous work robust saliency (ROSA) shuffles the pre-segmented superpixels and then refines the coarse saliency map by the densely connected conditional random field (CRF). Different from ROSA that relies on various pre- and post-processings, this paper proposes a light-weight Learnable Noise (LeNo) to defend adversarial attacks for SOD models. LeNo preserves accuracy of SOD models on both adversarial and clean images, as well as inference speed. In general, LeNo consists of a simple shallow noise and noise estimation that embedded in the encoder and decoder of arbitrary SOD networks respectively. Inspired by the center prior of human visual attention mechanism, we initialize the shallow noise with a cross-shaped gaussian distribution for better defense against adversarial attacks. Instead of adding additional network components for post-processing, the proposed noise estimation modifies only one channel of the decoder. With the deeply-supervised noise-decoupled training on state-of-the-art RGB and RGB-D SOD networks, LeNo outperforms previous works not only on adversarial images but also on clean images, which contributes stronger robustness for SOD. Our code is available at https://github.com/ssecv/LeNo.

arXiv:2210.15365

Li3DeTr: A LiDAR based 3D Detection Transformer

Inspired by recent advances in vision transformers for object detection, we propose Li3DeTr, an end-to-end LiDAR based 3D Detection Transformer for autonomous driving, that inputs LiDAR point clouds and regresses 3D bounding boxes. The LiDAR local and global features are encoded using sparse convolution and multi-scale deformable attention respectively. In the decoder head, firstly, in the novel Li3DeTr cross-attention block, we link the LiDAR global features to 3D predictions leveraging the sparse set of object queries learnt from the data. Secondly, the object query interactions are formulated using multi-head self-attention. Finally, the decoder layer is repeated $L_{dec}$ number of times to refine the object queries. Inspired by DETR, we employ set-to-set loss to train the Li3DeTr network. Without bells and whistles, the Li3DeTr network achieves 61.3% mAP and 67.6% NDS surpassing the state-of-the-art methods with non-maximum suppression (NMS) on the nuScenes dataset and it also achieves competitive performance on the KITTI dataset. We also employ knowledge distillation (KD) using a teacher and student model that slightly improves the performance of our network.

arXiv:2210.15156공식 코드 · Papers with Code

Towards Complex Backgrounds: A Unified Difference-Aware Decoder for Binary Segmentation

Binary segmentation is used to distinguish objects of interest from background, and is an active area of convolutional encoder-decoder network research. The current decoders are designed for specific objects based on the common backbones as the encoders, but cannot deal with complex backgrounds. Inspired by the way human eyes detect objects of interest, a new unified dual-branch decoder paradigm named the difference-aware decoder is proposed in this paper to explore the difference between the foreground and the background and separate the objects of interest in optical images. The difference-aware decoder imitates the human eye in three stages using the multi-level features output by the encoder. In Stage A, the first branch decoder of the difference-aware decoder is used to obtain a guide map. The highest-level features are enhanced with a novel field expansion module and a dual residual attention module, and are combined with the lowest-level features to obtain the guide map. In Stage B, the other branch decoder adopts a middle feature fusion module to make trade-offs between textural details and semantic information and generate background-aware features. In Stage C, the proposed difference-aware extractor, consisting of a difference guidance model and a difference enhancement module, fuses the guide map from Stage A and the background-aware features from Stage B, to enlarge the differences between the foreground and the background and output a final detection result. The results demonstrate that the difference-aware decoder can achieve a higher accuracy than the other state-of-the-art binary segmentation methods for these tasks.

arXiv:2210.15028

FaD-VLP: Fashion Vision-and-Language Pre-training towards Unified Retrieval and Captioning

Multimodal tasks in the fashion domain have significant potential for e-commerce, but involve challenging vision-and-language learning problems - e.g., retrieving a fashion item given a reference image plus text feedback from a user. Prior works on multimodal fashion tasks have either been limited by the data in individual benchmarks, or have leveraged generic vision-and-language pre-training but have not taken advantage of the characteristics of fashion data. Additionally, these works have mainly been restricted to multimodal understanding tasks. To address these gaps, we make two key contributions. First, we propose a novel fashion-specific pre-training framework based on weakly-supervised triplets constructed from fashion image-text pairs. We show the triplet-based tasks are an effective addition to standard multimodal pre-training tasks. Second, we propose a flexible decoder-based model architecture capable of both fashion retrieval and captioning tasks. Together, our model design and pre-training approach are competitive on a diverse set of fashion tasks, including cross-modal retrieval, image retrieval with text feedback, image captioning, relative image captioning, and multimodal categorization.

arXiv:2210.14725

Linguistic-Enhanced Transformer with CTC Embedding for Speech Recognition

The recent emergence of joint CTC-Attention model shows significant improvement in automatic speech recognition (ASR). The improvement largely lies in the modeling of linguistic information by decoder. The decoder joint-optimized with an acoustic encoder renders the language model from ground-truth sequences in an auto-regressive manner during training. However, the training corpus of the decoder is limited to the speech transcriptions, which is far less than the corpus needed to train an acceptable language model. This leads to poor robustness of decoder. To alleviate this problem, we propose linguistic-enhanced transformer, which introduces refined CTC information to decoder during training process, so that the decoder can be more robust. Our experiments on AISHELL-1 speech corpus show that the character error rate (CER) is relatively reduced by up to 7%. We also find that in joint CTC-Attention ASR model, decoder is more sensitive to linguistic information than acoustic information.

arXiv:2210.13761

Streaming Parrotron for on-device speech-to-speech conversion

We present a fully on-device streaming Speech2Speech conversion model that normalizes a given input speech directly to synthesized output speech. Deploying such a model on mobile devices pose significant challenges in terms of memory footprint and computation requirements. We present a streaming-based approach to produce an acceptable delay, with minimal loss in speech conversion quality, when compared to a reference state of the art non-streaming approach. Our method consists of first streaming the encoder in real time while the speaker is speaking. Then, as soon as the speaker stops speaking, we run the spectrogram decoder in streaming mode along the side of a streaming vocoder to generate output speech. To achieve an acceptable delay-quality trade-off, we propose a novel hybrid approach for look-ahead in the encoder which combines a look-ahead feature stacker with a look-ahead self-attention. We show that our streaming approach is almost 2x faster than real time on the Pixel4 CPU.

arXiv:2210.14749공식 코드 · Papers with Code

Hybrid HMM Decoder For Convolutional Codes By Joint Trellis-Like Structure and Channel Prior

The anti-interference capability of wireless links is a physical layer problem for edge computing. Although convolutional codes have inherent error correction potential due to the redundancy introduced in the data, the performance of the convolutional code is drastically degraded due to multipath effects on the channel. In this paper, we propose the use of a Hidden Markov Model (HMM) for the reconstruction of convolutional codes and decoding by the Viterbi algorithm. Furthermore, to implement soft-decision decoding, the observation of HMM is replaced by Gaussian mixture models (GMM). Our method provides superior error correction potential than the standard method because the model parameters contain channel state information (CSI). We evaluated the performance of the method compared to standard Viterbi decoding by numerical simulation. In the multipath channel, the hybrid HMM decoder can achieve a performance gain of 4.7 dB and 2 dB when using hard-decision and soft-decision decoding, respectively. The HMM decoder also achieves significant performance gains for the RSC code, suggesting that the method could be extended to turbo codes.

arXiv:2210.14742공식 코드 · Papers with Code

Monotonic segmental attention for automatic speech recognition

We introduce a novel segmental-attention model for automatic speech recognition. We restrict the decoder attention to segments to avoid quadratic runtime of global attention, better generalize to long sequences, and eventually enable streaming. We directly compare global-attention and different segmental-attention modeling variants. We develop and compare two separate time-synchronous decoders, one specifically taking the segmental nature into account, yielding further improvements. Using time-synchronous decoding for segmental models is novel and a step towards streaming applications. Our experiments show the importance of a length model to predict the segment boundaries. The final best segmental-attention model using segmental decoding performs better than global-attention, in contrast to other monotonic attention approaches in the literature. Further, we observe that the segmental model generalizes much better to long sequences of up to several minutes.

arXiv:2210.14391

Can Transformer Attention Spread Give Insights Into Uncertainty of Detected and Tracked Objects?

Transformers have recently been utilized to perform object detection and tracking in the context of autonomous driving. One unique characteristic of these models is that attention weights are computed in each forward pass, giving insights into the model's interior, in particular, which part of the input data it deemed interesting for the given task. Such an attention matrix with the input grid is available for each detected (or tracked) object in every transformer decoder layer. In this work, we investigate the distribution of these attention weights: How do they change through the decoder layers and through the lifetime of a track? Can they be used to infer additional information about an object, such as a detection uncertainty? Especially in unstructured environments, or environments that were not common during training, a reliable measure of detection uncertainty is crucial to decide whether the system can still be trusted or not.

arXiv:2210.14231

NAS-PRNet: Neural Architecture Search generated Phase Retrieval Net for Off-axis Quantitative Phase Imaging

Single neural networks have achieved simultaneous phase retrieval with aberration compensation and phase unwrapping in off-axis Quantitative Phase Imaging (QPI). However, when designing the phase retrieval neural network architecture, the trade-off between computation latency and accuracy has been largely neglected. Here, we propose Neural Architecture Search (NAS) generated Phase Retrieval Net (NAS-PRNet), which is an encoder-decoder style neural network, automatically found from a large neural network architecture search space. The NAS scheme in NAS-PRNet is modified from SparseMask, in which the learning of skip connections between the encoder and the decoder is formulated as a differentiable NAS problem, and the gradient decent is applied to efficiently search the optimal skip connections. Using MobileNet-v2 as the encoder and a synthesized loss that incorporates phase reconstruction and network sparsity losses, NAS-PRNet has realized fast and accurate phase retrieval of biological cells. When tested on a cell dataset, NAS-PRNet has achieved a Peak Signal-to-Noise Ratio (PSNR) of 36.1 dB, outperforming the widely used U-Net and original SparseMask-generated neural network. Notably, the computation latency of NAS-PRNet is only 31 ms which is 12 times less than U-Net. Moreover, the connectivity scheme in NAS-PRNet, identified from one off-axis QPI system, can be well fitted to another with different fringe patterns.

arXiv:2210.14593

Scaling Laws Beyond Backpropagation

Alternatives to backpropagation have long been studied to better understand how biological brains may learn. Recently, they have also garnered interest as a way to train neural networks more efficiently. By relaxing constraints inherent to backpropagation (e.g., symmetric feedforward and feedback weights, sequential updates), these methods enable promising prospects, such as local learning. However, the tradeoffs between different methods in terms of final task performance, convergence speed, and ultimately compute and data requirements are rarely outlined. In this work, we use scaling laws to study the ability of Direct Feedback Alignment~(DFA) to train causal decoder-only Transformers efficiently. Scaling laws provide an overview of the tradeoffs implied by a modeling decision, up to extrapolating how it might transfer to increasingly large models. We find that DFA fails to offer more efficient scaling than backpropagation: there is never a regime for which the degradation in loss incurred by using DFA is worth the potential reduction in compute budget. Our finding comes at variance with previous beliefs in the alternative training methods community, and highlights the need for holistic empirical approaches to better understand modeling decisions.

arXiv:2210.14530공식 코드 · Papers with Code

RGB-T Semantic Segmentation with Location, Activation, and Sharpening

Semantic segmentation is important for scene understanding. To address the scenes of adverse illumination conditions of natural images, thermal infrared (TIR) images are introduced. Most existing RGB-T semantic segmentation methods follow three cross-modal fusion paradigms, i.e. encoder fusion, decoder fusion, and feature fusion. Some methods, unfortunately, ignore the properties of RGB and TIR features or the properties of features at different levels. In this paper, we propose a novel feature fusion-based network for RGB-T semantic segmentation, named \emph{LASNet}, which follows three steps of location, activation, and sharpening. The highlight of LASNet is that we fully consider the characteristics of cross-modal features at different levels, and accordingly propose three specific modules for better segmentation. Concretely, we propose a Collaborative Location Module (CLM) for high-level semantic features, aiming to locate all potential objects. We propose a Complementary Activation Module for middle-level features, aiming to activate exact regions of different objects. We propose an Edge Sharpening Module (ESM) for low-level texture features, aiming to sharpen the edges of objects. Furthermore, in the training phase, we attach a location supervision and an edge supervision after CLM and ESM, respectively, and impose two semantic supervisions in the decoder part to facilitate network convergence. Experimental results on two public datasets demonstrate that the superiority of our LASNet over relevant state-of-the-art methods. The code and results of our method are available at https://github.com/MathLee/LASNet.

arXiv:2210.13803

Adapitch: Adaption Multi-Speaker Text-to-Speech Conditioned on Pitch Disentangling with Untranscribed Data

In this paper, we proposed Adapitch, a multi-speaker TTS method that makes adaptation of the supervised module with untranscribed data. We design two self supervised modules to train the text encoder and mel decoder separately with untranscribed data to enhance the representation of text and mel. To better handle the prosody information in a synthesized voice, a supervised TTS module is designed conditioned on content disentangling of pitch, text, and speaker. The training phase was separated into two parts, pretrained and fixed the text encoder and mel decoder with unsupervised mode, then the supervised mode on the disentanglement of TTS. Experiment results show that the Adaptich achieved much better quality than baseline methods.

arXiv:2210.13774공식 코드 · Papers with Code

From Points to Functions: Infinite-dimensional Representations in Diffusion Models

Diffusion-based generative models learn to iteratively transfer unstructured noise to a complex target distribution as opposed to Generative Adversarial Networks (GANs) or the decoder of Variational Autoencoders (VAEs) which produce samples from the target distribution in a single step. Thus, in diffusion models every sample is naturally connected to a random trajectory which is a solution to a learned stochastic differential equation (SDE). Generative models are only concerned with the final state of this trajectory that delivers samples from the desired distribution. Abstreiter et. al showed that these stochastic trajectories can be seen as continuous filters that wash out information along the way. Consequently, it is reasonable to ask if there is an intermediate time step at which the preserved information is optimal for a given downstream task. In this work, we show that a combination of information content from different time steps gives a strictly better representation for the downstream task. We introduce an attention and recurrence based modules that ``learn to mix'' information content of various time-steps such that the resultant representation leads to superior performance in downstream tasks.

arXiv:2210.14147

Food Ingredients Recognition through Multi-label Learning

The ability to recognize various food-items in a generic food plate is a key determinant for an automated diet assessment system. This study motivates the need for automated diet assessment and proposes a framework to achieve this. Within this framework, we focus on one of the core functionalities to visually recognize various ingredients. To this end, we employed a deep multi-label learning approach and evaluated several state-of-the-art neural networks for their ability to detect an arbitrary number of ingredients in a dish image. The models evaluated in this work follow a definite meta-structure, consisting of an encoder and a decoder component. Two distinct decoding schemes, one based on global average pooling and the other on attention mechanism, are evaluated and benchmarked. Whereas for encoding, several well-known architectures, including DenseNet, EfficientNet, MobileNet, Inception and Xception, were employed. We present promising preliminary results for deep learning-based ingredients detection, using a challenging dataset, Nutrition5K, and establish a strong baseline for future explorations.

arXiv:2210.13984공식 코드 · Papers with Code

Inferring Past Human Actions in Homes with Abductive Reasoning

Abductive reasoning aims to make the most likely inference for a given set of incomplete observations. In this paper, we introduce "Abductive Past Action Inference", a novel research task aimed at identifying the past actions performed by individuals within homes to reach specific states captured in a single image, using abductive inference. The research explores three key abductive inference problems: past action set prediction, past action sequence prediction, and abductive past action verification. We introduce several models tailored for abductive past action inference, including a relational graph neural network, a relational bilinear pooling model, and a relational transformer model. Notably, the newly proposed object-relational bilinear graph encoder-decoder (BiGED) model emerges as the most effective among all methods evaluated, demonstrating good proficiency in handling the intricacies of the Action Genome dataset. The contributions of this research significantly advance the ability of deep learning models to reason about current scene evidence and make highly plausible inferences about past human actions. This advancement enables a deeper understanding of events and behaviors, which can enhance decision-making and improve system capabilities across various real-world applications such as Human-Robot Interaction and Elderly Care and Health Monitoring. Code and data available at https://github.com/LUNAProject22/AAR

arXiv:2210.13821공식 코드 · Papers with Code

Salient Object Detection via Dynamic Scale Routing

Recent research advances in salient object detection (SOD) could largely be attributed to ever-stronger multi-scale feature representation empowered by the deep learning technologies. The existing SOD deep models extract multi-scale features via the off-the-shelf encoders and combine them smartly via various delicate decoders. However, the kernel sizes in this commonly-used thread are usually "fixed". In our new experiments, we have observed that kernels of small size are preferable in scenarios containing tiny salient objects. In contrast, large kernel sizes could perform better for images with large salient objects. Inspired by this observation, we advocate the "dynamic" scale routing (as a brand-new idea) in this paper. It will result in a generic plug-in that could directly fit the existing feature backbone. This paper's key technical innovations are two-fold. First, instead of using the vanilla convolution with fixed kernel sizes for the encoder design, we propose the dynamic pyramid convolution (DPConv), which dynamically selects the best-suited kernel sizes w.r.t. the given input. Second, we provide a self-adaptive bidirectional decoder design to accommodate the DPConv-based encoder best. The most significant highlight is its capability of routing between feature scales and their dynamic collection, making the inference process scale-aware. As a result, this paper continues to enhance the current SOTA performance. Both the code and dataset are publicly available at https://github.com/wuzhenyubuaa/DPNet.