PyTorchKR Papers

최신 논문

총 5,360편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2211.01519

SLICER: Learning universal audio representations using low-resource self-supervised pre-training

We present a new Self-Supervised Learning (SSL) approach to pre-train encoders on unlabeled audio data that reduces the need for large amounts of labeled data for audio and speech classification. Our primary aim is to learn audio representations that can generalize across a large variety of speech and non-speech tasks in a low-resource un-labeled audio pre-training setting. Inspired by the recent success of clustering and contrasting learning paradigms for SSL-based speech representation learning, we propose SLICER (Symmetrical Learning of Instance and Cluster-level Efficient Representations), which brings together the best of both clustering and contrasting learning paradigms. We use a symmetric loss between latent representations from student and teacher encoders and simultaneously solve instance and cluster-level contrastive learning tasks. We obtain cluster representations online by just projecting the input spectrogram into an output subspace with dimensions equal to the number of clusters. In addition, we propose a novel mel-spectrogram augmentation procedure, k-mix, based on mixup, which does not require labels and aids unsupervised representation learning for audio. Overall, SLICER achieves state-of-the-art results on the LAPE Benchmark \cite{9868132}, significantly outperforming DeLoRes-M and other prior approaches, which are pre-trained on $10\times$ larger of unsupervised data. We will make all our codes available on GitHub.

arXiv:2211.01515공식 코드 · Papers with Code

MAST: Multiscale Audio Spectrogram Transformers

We present Multiscale Audio Spectrogram Transformer (MAST) for audio classification, which brings the concept of multiscale feature hierarchies to the Audio Spectrogram Transformer (AST). Given an input audio spectrogram, we first patchify and project it into an initial temporal resolution and embedding dimension, post which the multiple stages in MAST progressively expand the embedding dimension while reducing the temporal resolution of the input. We use a pyramid structure that allows early layers of MAST operating at a high temporal resolution but low embedding space to model simple low-level acoustic information and deeper temporally coarse layers to model high-level acoustic information with high-dimensional embeddings. We also extend our approach to present a new Self-Supervised Learning (SSL) method called SS-MAST, which calculates a symmetric contrastive loss between latent representations from a student and a teacher encoder, leveraging patch-drop, a novel audio augmentation approach that we introduce. In practice, MAST significantly outperforms AST by an average accuracy of 3.4% across 8 speech and non-speech tasks from the LAPE Benchmark, achieving state-of-the-art results on keyword spotting in Speech Commands. Additionally, our proposed SS-MAST achieves an absolute average improvement of 2.6% over the previously proposed SSAST.

arXiv:2211.00854

More Speaking or More Speakers?

Self-training (ST) and self-supervised learning (SSL) methods have demonstrated strong improvements in automatic speech recognition (ASR). In spite of these advances, to the best of our knowledge, there is no analysis of how the composition of the labelled and unlabelled datasets used in these methods affects the results. In this work we aim to analyse the effect of number of speakers in the training data on a recent SSL algorithm (wav2vec 2.0), and a recent ST algorithm (slimIPL). We perform a systematic analysis on both labeled and unlabeled data by varying the number of speakers while keeping the number of hours fixed and vice versa. Our findings suggest that SSL requires a large amount of unlabeled data to produce high accuracy results, while ST requires a sufficient number of speakers in the labelled data, especially in the low-regime setting. In this manner these two approaches improve supervised learning in different regimes of data composition.

arXiv:2211.01246공식 코드 · Papers with Code

data2vec-aqc: Search for the right Teaching Assistant in the Teacher-Student training setup

In this paper, we propose a new Self-Supervised Learning (SSL) algorithm called data2vec-aqc, for speech representation learning from unlabeled speech data. Our goal is to improve SSL for speech in domains where both unlabeled and labeled data are limited. Building on the recently introduced data2vec, we introduce additional modules to the data2vec framework that leverage the benefit of data augmentations, quantized representations, and clustering. The interaction between these modules helps solve the cross-contrastive loss as an additional self-supervised objective. data2vec-aqc achieves up to 14.1% and 20.9% relative WER improvement over the existing state-of-the-art data2vec system over the test-clean and test-other sets, respectively of LibriSpeech, without the use of any language model (LM). Our proposed model also achieves up to 17.8\% relative WER gains over the baseline data2vec when fine-tuned on a subset of the Switchboard dataset. Code: https://github.com/Speech-Lab-IITM/data2vec-aqc.

arXiv:2211.01244공식 코드 · Papers with Code

EquiMod: An Equivariance Module to Improve Self-Supervised Learning

Self-supervised visual representation methods are closing the gap with supervised learning performance. These methods rely on maximizing the similarity between embeddings of related synthetic inputs created through data augmentations. This can be seen as a task that encourages embeddings to leave out factors modified by these augmentations, i.e. to be invariant to them. However, this only considers one side of the trade-off in the choice of the augmentations: they need to strongly modify the images to avoid simple solution shortcut learning (e.g. using only color histograms), but on the other hand, augmentations-related information may be lacking in the representations for some downstream tasks (e.g. color is important for birds and flower classification). Few recent works proposed to mitigate the problem of using only an invariance task by exploring some form of equivariance to augmentations. This has been performed by learning additional embeddings space(s), where some augmentation(s) cause embeddings to differ, yet in a non-controlled way. In this work, we introduce EquiMod a generic equivariance module that structures the learned latent space, in the sense that our module learns to predict the displacement in the embedding space caused by the augmentations. We show that applying that module to state-of-the-art invariance models, such as SimCLR and BYOL, increases the performances on CIFAR10 and ImageNet datasets. Moreover, while our model could collapse to a trivial equivariance, i.e. invariance, we observe that it instead automatically learns to keep some augmentations-related information beneficial to the representations.

arXiv:2211.01165

RegCLR: A Self-Supervised Framework for Tabular Representation Learning in the Wild

Recent advances in self-supervised learning (SSL) using large models to learn visual representations from natural images are rapidly closing the gap between the results produced by fully supervised learning and those produced by SSL on downstream vision tasks. Inspired by this advancement and primarily motivated by the emergence of tabular and structured document image applications, we investigate which self-supervised pretraining objectives, architectures, and fine-tuning strategies are most effective. To address these questions, we introduce RegCLR, a new self-supervised framework that combines contrastive and regularized methods and is compatible with the standard Vision Transformer architecture. Then, RegCLR is instantiated by integrating masked autoencoders as a representative example of a contrastive method and enhanced Barlow Twins as a representative example of a regularized method with configurable input image augmentations in both branches. Several real-world table recognition scenarios (e.g., extracting tables from document images), ranging from standard Word and Latex documents to even more challenging electronic health records (EHR) computer screen images, have been shown to benefit greatly from the representations learned from this new framework, with detection average-precision (AP) improving relatively by 4.8% for Table, 11.8% for Column, and 11.1% for GUI objects over a previous fully supervised baseline on real-world EHR screen images.

arXiv:2211.00482

Adapting self-supervised models to multi-talker speech recognition using speaker embeddings

Self-supervised learning (SSL) methods which learn representations of data without explicit supervision have gained popularity in speech-processing tasks, particularly for single-talker applications. However, these models often have degraded performance for multi-talker scenarios -- possibly due to the domain mismatch -- which severely limits their use for such applications. In this paper, we investigate the adaptation of upstream SSL models to the multi-talker automatic speech recognition (ASR) task under two conditions. First, when segmented utterances are given, we show that adding a target speaker extraction (TSE) module based on enrollment embeddings is complementary to mixture-aware pre-training. Second, for unsegmented mixtures, we propose a novel joint speaker modeling (JSM) approach, which aggregates information from all speakers in the mixture through their embeddings. With controlled experiments on Libri2Mix, we show that using speaker embeddings provides relative WER improvements of 9.1% and 42.1% over strong baselines for the segmented and unsegmented cases, respectively. We also demonstrate the effectiveness of our models for real conversational mixtures through experiments on the AMI dataset.

arXiv:2211.00342

Investigating Content-Aware Neural Text-To-Speech MOS Prediction Using Prosodic and Linguistic Features

Current state-of-the-art methods for automatic synthetic speech evaluation are based on MOS prediction neural models. Such MOS prediction models include MOSNet and LDNet that use spectral features as input, and SSL-MOS that relies on a pretrained self-supervised learning model that directly uses the speech signal as input. In modern high-quality neural TTS systems, prosodic appropriateness with regard to the spoken content is a decisive factor for speech naturalness. For this reason, we propose to include prosodic and linguistic features as additional inputs in MOS prediction systems, and evaluate their impact on the prediction outcome. We consider phoneme level F0 and duration features as prosodic inputs, as well as Tacotron encoder outputs, POS tags and BERT embeddings as higher-level linguistic inputs. All MOS prediction systems are trained on SOMOS, a neural TTS-only dataset with crowdsourced naturalness MOS evaluations. Results show that the proposed additional features are beneficial in the MOS prediction task, by improving the predicted MOS scores' correlation with the ground truths, both at utterance-level and system-level predictions.

arXiv:2211.00527

Self-Supervised Learning with Limited Labeled Data for Prostate Cancer Detection in High Frequency Ultrasound

Deep learning-based analysis of high-frequency, high-resolution micro-ultrasound data shows great promise for prostate cancer detection. Previous approaches to analysis of ultrasound data largely follow a supervised learning paradigm. Ground truth labels for ultrasound images used for training deep networks often include coarse annotations generated from the histopathological analysis of tissue samples obtained via biopsy. This creates inherent limitations on the availability and quality of labeled data, posing major challenges to the success of supervised learning methods. On the other hand, unlabeled prostate ultrasound data are more abundant. In this work, we successfully apply self-supervised representation learning to micro-ultrasound data. Using ultrasound data from 1028 biopsy cores of 391 subjects obtained in two clinical centres, we demonstrate that feature representations learnt with this method can be used to classify cancer from non-cancer tissue, obtaining an AUROC score of 91% on an independent test set. To the best of our knowledge, this is the first successful end-to-end self-supervised learning approach for prostate cancer detection using ultrasound data. Our method outperforms baseline supervised learning approaches, generalizes well between different data centers, and scale well in performance as more unlabeled data are added, making it a promising approach for future research using large volumes of unlabeled data.

arXiv:2211.00509

Self-Supervised Intensity-Event Stereo Matching

Event cameras are novel bio-inspired vision sensors that output pixel-level intensity changes in microsecond accuracy with a high dynamic range and low power consumption. Despite these advantages, event cameras cannot be directly applied to computational imaging tasks due to the inability to obtain high-quality intensity and events simultaneously. This paper aims to connect a standalone event camera and a modern intensity camera so that the applications can take advantage of both two sensors. We establish this connection through a multi-modal stereo matching task. We first convert events to a reconstructed image and extend the existing stereo networks to this multi-modality condition. We propose a self-supervised method to train the multi-modal stereo network without using ground truth disparity data. The structure loss calculated on image gradients is used to enable self-supervised learning on such multi-modal data. Exploiting the internal stereo constraint between views with different modalities, we introduce general stereo loss functions, including disparity cross-consistency loss and internal disparity loss, leading to improved performance and robustness compared to existing approaches. The experiments demonstrate the effectiveness of the proposed method, especially the proposed general stereo loss functions, on both synthetic and real datasets. At last, we shed light on employing the aligned events and intensity images in downstream tasks, e.g., video interpolation application.

arXiv:2211.00313

RGMIM: Region-Guided Masked Image Modeling for Learning Meaningful Representations from X-Ray Images

In this study, we propose a novel method called region-guided masked image modeling (RGMIM) for learning meaningful representations from X-ray images. Our method adopts a new masking strategy that utilizes organ mask information to identify valid regions for learning more meaningful representations. We conduct quantitative evaluations on an open lung X-ray image dataset as well as masking ratio hyperparameter studies. When using the entire training set, RGMIM outperformed other comparable methods, achieving a 0.962 lung disease detection accuracy. Specifically, RGMIM significantly improved performance in small data volumes, such as 5% and 10% of the training set compared to other methods. RGMIM can mask more valid regions, facilitating the learning of discriminative representations and the subsequent high-accuracy lung disease detection. RGMIM outperforms other state-of-the-art self-supervised learning methods in experiments, particularly when limited training data is used.

arXiv:2211.00288공식 코드 · Papers with Code

Self-supervised Character-to-Character Distillation for Text Recognition

When handling complicated text images (e.g., irregular structures, low resolution, heavy occlusion, and uneven illumination), existing supervised text recognition methods are data-hungry. Although these methods employ large-scale synthetic text images to reduce the dependence on annotated real images, the domain gap still limits the recognition performance. Therefore, exploring the robust text feature representations on unlabeled real images by self-supervised learning is a good solution. However, existing self-supervised text recognition methods conduct sequence-to-sequence representation learning by roughly splitting the visual features along the horizontal axis, which limits the flexibility of the augmentations, as large geometric-based augmentations may lead to sequence-to-sequence feature inconsistency. Motivated by this, we propose a novel self-supervised Character-to-Character Distillation method, CCD, which enables versatile augmentations to facilitate general text representation learning. Specifically, we delineate the character structures of unlabeled real images by designing a self-supervised character segmentation module. Following this, CCD easily enriches the diversity of local characters while keeping their pairwise alignment under flexible augmentations, using the transformation matrix between two augmented views from images. Experiments demonstrate that CCD achieves state-of-the-art results, with average performance gains of 1.38% in text recognition, 1.7% in text segmentation, 0.24 dB (PSNR) and 0.0321 (SSIM) in text super-resolution. Code is available at https://github.com/TongkunGuan/CCD.

arXiv:2210.16871공식 코드 · Papers with Code

Improved acoustic-to-articulatory inversion using representations from pretrained self-supervised learning models

In this work, we investigate the effectiveness of pretrained Self-Supervised Learning (SSL) features for learning the mapping for acoustic to articulatory inversion (AAI). Signal processing-based acoustic features such as MFCCs have been predominantly used for the AAI task with deep neural networks. With SSL features working well for various other speech tasks such as speech recognition, emotion classification, etc., we experiment with its efficacy for AAI. We train on SSL features with transformer neural networks-based AAI models of 3 different model complexities and compare its performance with MFCCs in subject-specific (SS), pooled and fine-tuned (FT) configurations with data from 10 subjects, and evaluate with correlation coefficient (CC) score on the unseen sentence test set. We find that acoustic feature reconstruction objective-based SSL features such as TERA and DeCoAR work well for AAI, with SS CCs of these SSL features reaching close to the best FT CCs of MFCC. We also find the results consistent across different model sizes.

arXiv:2210.17456

Audio-Visual Speech Enhancement and Separation by Utilizing Multi-Modal Self-Supervised Embeddings

AV-HuBERT, a multi-modal self-supervised learning model, has been shown to be effective for categorical problems such as automatic speech recognition and lip-reading. This suggests that useful audio-visual speech representations can be obtained via utilizing multi-modal self-supervised embeddings. Nevertheless, it is unclear if such representations can be generalized to solve real-world multi-modal AV regression tasks, such as audio-visual speech enhancement (AVSE) and audio-visual speech separation (AVSS). In this study, we leveraged the pre-trained AV-HuBERT model followed by an SE module for AVSE and AVSS. Comparative experimental results demonstrate that our proposed model performs better than the state-of-the-art AVSE and traditional audio-only SE models. In summary, our results confirm the effectiveness of our proposed model for the AVSS task with proper fine-tuning strategies, demonstrating that multi-modal self-supervised embeddings obtained from AV-HuBERT can be generalized to audio-visual regression tasks.

arXiv:2210.16659

Learning Dependencies of Discrete Speech Representations with Neural Hidden Markov Models

While discrete latent variable models have had great success in self-supervised learning, most models assume that frames are independent. Due to the segmental nature of phonemes in speech perception, modeling dependencies among latent variables at the frame level can potentially improve the learned representations on phonetic-related tasks. In this work, we assume Markovian dependencies among latent variables, and propose to learn speech representations with neural hidden Markov models. Our general framework allows us to compare to self-supervised models that assume independence, while keeping the number of parameters fixed. The added dependencies improve the accessibility of phonetic information, phonetic segmentation, and the cluster purity of phones, showcasing the benefit of the assumed dependencies.

arXiv:2210.16365

Elastic Weight Consolidation Improves the Robustness of Self-Supervised Learning Methods under Transfer

Self-supervised representation learning (SSL) methods provide an effective label-free initial condition for fine-tuning downstream tasks. However, in numerous realistic scenarios, the downstream task might be biased with respect to the target label distribution. This in turn moves the learned fine-tuned model posterior away from the initial (label) bias-free self-supervised model posterior. In this work, we re-interpret SSL fine-tuning under the lens of Bayesian continual learning and consider regularization through the Elastic Weight Consolidation (EWC) framework. We demonstrate that self-regularization against an initial SSL backbone improves worst sub-group performance in Waterbirds by 5% and Celeb-A by 2% when using the ViT-B/16 architecture. Furthermore, to help simplify the use of EWC with SSL, we pre-compute and publicly release the Fisher Information Matrix (FIM), evaluated with 10,000 ImageNet-1K variates evaluated on large modern SSL architectures including ViT-B/16 and ResNet50 trained with DINO.

arXiv:2210.17400공식 코드 · Papers with Code

Max Pooling with Vision Transformers reconciles class and shape in weakly supervised semantic segmentation

Weakly Supervised Semantic Segmentation (WSSS) research has explored many directions to improve the typical pipeline CNN plus class activation maps (CAM) plus refinements, given the image-class label as the only supervision. Though the gap with the fully supervised methods is reduced, further abating the spread seems unlikely within this framework. On the other hand, WSSS methods based on Vision Transformers (ViT) have not yet explored valid alternatives to CAM. ViT features have been shown to retain a scene layout, and object boundaries in self-supervised learning. To confirm these findings, we prove that the advantages of transformers in self-supervised methods are further strengthened by Global Max Pooling (GMP), which can leverage patch features to negotiate pixel-label probability with class probability. This work proposes a new WSSS method dubbed ViT-PCM (ViT Patch-Class Mapping), not based on CAM. The end-to-end presented network learns with a single optimization process, refined shape and proper localization for segmentation masks. Our model outperforms the state-of-the-art on baseline pseudo-masks (BPM), where we achieve $69.3\%$ mIoU on PascalVOC 2012 $val$ set. We show that our approach has the least set of parameters, though obtaining higher accuracy than all other approaches. In a sentence, quantitative and qualitative results of our method reveal that ViT-PCM is an excellent alternative to CNN-CAM based architectures.

arXiv:2210.16870

A simple, efficient and scalable contrastive masked autoencoder for learning visual representations

We introduce CAN, a simple, efficient and scalable method for self-supervised learning of visual representations. Our framework is a minimal and conceptually clean synthesis of (C) contrastive learning, (A) masked autoencoders, and (N) the noise prediction approach used in diffusion models. The learning mechanisms are complementary to one another: contrastive learning shapes the embedding space across a batch of image samples; masked autoencoders focus on reconstruction of the low-frequency spatial correlations in a single image sample; and noise prediction encourages the reconstruction of the high-frequency components of an image. The combined approach results in a robust, scalable and simple-to-implement algorithm. The training process is symmetric, with 50% of patches in both views being masked at random, yielding a considerable efficiency improvement over prior contrastive learning methods. Extensive empirical studies demonstrate that CAN achieves strong downstream performance under both linear and finetuning evaluations on transfer learning and robustness tasks. CAN outperforms MAE and SimCLR when pre-training on ImageNet, but is especially useful for pre-training on larger uncurated datasets such as JFT-300M: for linear probe on ImageNet, CAN achieves 75.4% compared to 73.4% for SimCLR and 64.1% for MAE. The finetuned performance on ImageNet of our ViT-L model is 86.1%, compared to 85.5% for SimCLR, and 85.4% for MAE. The overall FLOPs load of SimCLR is 70% higher than CAN for ViT-L models.

arXiv:2210.16776

Saliency Can Be All You Need In Contrastive Self-Supervised Learning

We propose an augmentation policy for Contrastive Self-Supervised Learning (SSL) in the form of an already established Salient Image Segmentation technique entitled Global Contrast based Salient Region Detection. This detection technique, which had been devised for unrelated Computer Vision tasks, was empirically observed to play the role of an augmentation facilitator within the SSL protocol. This observation is rooted in our practical attempts to learn, by SSL-fashion, aerial imagery of solar panels, which exhibit challenging boundary patterns. Upon the successful integration of this technique on our problem domain, we formulated a generalized procedure and conducted a comprehensive, systematic performance assessment with various Contrastive SSL algorithms subject to standard augmentation techniques. This evaluation, which was conducted across multiple datasets, indicated that the proposed technique indeed contributes to SSL. We hypothesize whether salient image segmentation may suffice as the only augmentation policy in Contrastive SSL when treating downstream segmentation tasks.

arXiv:2210.16476

Pair DETR: Contrastive Learning Speeds Up DETR Training

The DETR object detection approach applies the transformer encoder and decoder architecture to detect objects and achieves promising performance. In this paper, we present a simple approach to address the main problem of DETR, the slow convergence, by using representation learning technique. In this approach, we detect an object bounding box as a pair of keypoints, the top-left corner and the center, using two decoders. By detecting objects as paired keypoints, the model builds up a joint classification and pair association on the output queries from two decoders. For the pair association we propose utilizing contrastive self-supervised learning algorithm without requiring specialized architecture. Experimental results on MS COCO dataset show that Pair DETR can converge at least 10x faster than original DETR and 1.5x faster than Conditional DETR during training, while having consistently higher Average Precision scores.