PyTorchKR Papers

최신 논문

총 1,512편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2308.08669공식 코드 · Papers with Code

SkinDistilViT: Lightweight Vision Transformer for Skin Lesion Classification

Skin cancer is a treatable disease if discovered early. We provide a production-specific solution to the skin cancer classification problem that matches human performance in melanoma identification by training a vision transformer on melanoma medical images annotated by experts. Since inference cost, both time and memory wise is important in practice, we employ knowledge distillation to obtain a model that retains 98.33% of the teacher's balanced multi-class accuracy, at a fraction of the cost. Memory-wise, our model is 49.60% smaller than the teacher. Time-wise, our solution is 69.25% faster on GPU and 97.96% faster on CPU. By adding classification heads at each level of the transformer and employing a cascading distillation process, we improve the balanced multi-class accuracy of the base model by 2.1%, while creating a range of models of various sizes but comparable performance. We provide the code at https://github.com/Longman-Stan/SkinDistilVit.

arXiv:2308.09544공식 코드 · Papers with Code

Adapt Your Teacher: Improving Knowledge Distillation for Exemplar-free Continual Learning

In this work, we investigate exemplar-free class incremental learning (CIL) with knowledge distillation (KD) as a regularization strategy, aiming to prevent forgetting. KD-based methods are successfully used in CIL, but they often struggle to regularize the model without access to exemplars of the training data from previous tasks. Our analysis reveals that this issue originates from substantial representation shifts in the teacher network when dealing with out-of-distribution data. This causes large errors in the KD loss component, leading to performance degradation in CIL models. Inspired by recent test-time adaptation methods, we introduce Teacher Adaptation (TA), a method that concurrently updates the teacher and the main models during incremental training. Our method seamlessly integrates with KD-based CIL approaches and allows for consistent enhancement of their performance across multiple exemplar-free CIL benchmarks. The source code for our method is available at https://github.com/fszatkowski/cl-teacher-adaptation.

arXiv:2308.06453공식 코드 · Papers with Code

Multi-Label Knowledge Distillation

Existing knowledge distillation methods typically work by imparting the knowledge of output logits or intermediate feature maps from the teacher network to the student network, which is very successful in multi-class single-label learning. However, these methods can hardly be extended to the multi-label learning scenario, where each instance is associated with multiple semantic labels, because the prediction probabilities do not sum to one and feature maps of the whole example may ignore minor classes in such a scenario. In this paper, we propose a novel multi-label knowledge distillation method. On one hand, it exploits the informative semantic knowledge from the logits by dividing the multi-label learning problem into a set of binary classification problems; on the other hand, it enhances the distinctiveness of the learned feature representations by leveraging the structural information of label-wise embeddings. Experimental results on multiple benchmark datasets validate that the proposed method can avoid knowledge counteraction among labels, thus achieving superior performance against diverse comparing methods. Our code is available at: https://github.com/penghui-yang/L2D

arXiv:2308.06197공식 코드 · Papers with Code

Complex Facial Expression Recognition Using Deep Knowledge Distillation of Basic Features

Complex emotion recognition is a cognitive task that has so far eluded the same excellent performance of other tasks that are at or above the level of human cognition. Emotion recognition through facial expressions is particularly difficult due to the complexity of emotions expressed by the human face. For a machine to approach the same level of performance in complex facial expression recognition as a human, it may need to synthesise knowledge and understand new concepts in real-time, as humans do. Humans are able to learn new concepts using only few examples by distilling important information from memories. Inspired by human cognition and learning, we propose a novel continual learning method for complex facial expression recognition that can accurately recognise new compound expression classes using few training samples, by building on and retaining its knowledge of basic expression classes. In this work, we also use GradCAM visualisations to demonstrate the relationship between basic and compound facial expressions. Our method leverages this relationship through knowledge distillation and a novel Predictive Sorting Memory Replay, to achieve the current state-of-the-art in continual learning for complex facial expression recognition, with 74.28% Overall Accuracy on new classes. We also demonstrate that using continual learning for complex facial expression recognition achieves far better performance than non-continual learning methods, improving on state-of-the-art non-continual learning methods by 13.95%. Our work is also the first to apply few-shot learning to complex facial expression recognition, achieving the state-of-the-art with 100% accuracy using only a single training sample per class.

arXiv:2308.04779공식 코드 · Papers with Code

Multi-View Fusion and Distillation for Subgrade Distresses Detection based on 3D-GPR

The application of 3D ground-penetrating radar (3D-GPR) for subgrade distress detection has gained widespread popularity. To enhance the efficiency and accuracy of detection, pioneering studies have attempted to adopt automatic detection techniques, particularly deep learning. However, existing works typically rely on traditional 1D A-scan, 2D B-scan or 3D C-scan data of the GPR, resulting in either insufficient spatial information or high computational complexity. To address these challenges, we introduce a novel methodology for the subgrade distress detection task by leveraging the multi-view information from 3D-GPR data. Moreover, we construct a real multi-view image dataset derived from the original 3D-GPR data for the detection task, which provides richer spatial information compared to A-scan and B-scan data, while reducing computational complexity compared to C-scan data. Subsequently, we develop a novel \textbf{M}ulti-\textbf{V}iew \textbf{V}usion and \textbf{D}istillation framework, \textbf{GPR-MVFD}, specifically designed to optimally utilize the multi-view GPR dataset. This framework ingeniously incorporates multi-view distillation and attention-based fusion to facilitate significant feature extraction for subgrade distresses. In addition, a self-adaptive learning mechanism is adopted to stabilize the model training and prevent performance degeneration in each branch. Extensive experiments conducted on this new GPR benchmark demonstrate the effectiveness and efficiency of our proposed framework. Our framework outperforms not only the existing GPR baselines, but also the state-of-the-art methods in the fields of multi-view learning, multi-modal learning, and knowledge distillation. We will release the constructed multi-view GPR dataset with expert-annotated labels and the source codes of the proposed framework.

arXiv:2308.04070공식 코드 · Papers with Code

ConDistFL: Conditional Distillation for Federated Learning from Partially Annotated Data

Developing a generalized segmentation model capable of simultaneously delineating multiple organs and diseases is highly desirable. Federated learning (FL) is a key technology enabling the collaborative development of a model without exchanging training data. However, the limited access to fully annotated training data poses a major challenge to training generalizable models. We propose "ConDistFL", a framework to solve this problem by combining FL with knowledge distillation. Local models can extract the knowledge of unlabeled organs and tumors from partially annotated data from the global model with an adequately designed conditional probability representation. We validate our framework on four distinct partially annotated abdominal CT datasets from the MSD and KiTS19 challenges. The experimental results show that the proposed framework significantly outperforms FedAvg and FedOpt baselines. Moreover, the performance on an external test dataset demonstrates superior generalizability compared to models trained on each dataset separately. Our ablation study suggests that ConDistFL can perform well without frequent aggregation, reducing the communication cost of FL. Our implementation will be available at https://github.com/NVIDIA/NVFlare/tree/dev/research/condist-fl.

arXiv:2308.04243공식 코드 · Papers with Code

AICSD: Adaptive Inter-Class Similarity Distillation for Semantic Segmentation

In recent years, deep neural networks have achieved remarkable accuracy in computer vision tasks. With inference time being a crucial factor, particularly in dense prediction tasks such as semantic segmentation, knowledge distillation has emerged as a successful technique for improving the accuracy of lightweight student networks. The existing methods often neglect the information in channels and among different classes. To overcome these limitations, this paper proposes a novel method called Inter-Class Similarity Distillation (ICSD) for the purpose of knowledge distillation. The proposed method transfers high-order relations from the teacher network to the student network by independently computing intra-class distributions for each class from network outputs. This is followed by calculating inter-class similarity matrices for distillation using KL divergence between distributions of each pair of classes. To further improve the effectiveness of the proposed method, an Adaptive Loss Weighting (ALW) training strategy is proposed. Unlike existing methods, the ALW strategy gradually reduces the influence of the teacher network towards the end of training process to account for errors in teacher's predictions. Extensive experiments conducted on two well-known datasets for semantic segmentation, Cityscapes and Pascal VOC 2012, validate the effectiveness of the proposed method in terms of mIoU and pixel accuracy. The proposed method outperforms most of existing knowledge distillation methods as demonstrated by both quantitative and qualitative evaluations. Code is available at: https://github.com/AmirMansurian/AICSD

arXiv:2308.04061공식 코드 · Papers with Code

Enhancing Adversarial Robustness in Low-Label Regime via Adaptively Weighted Regularization and Knowledge Distillation

Adversarial robustness is a research area that has recently received a lot of attention in the quest for trustworthy artificial intelligence. However, recent works on adversarial robustness have focused on supervised learning where it is assumed that labeled data is plentiful. In this paper, we investigate semi-supervised adversarial training where labeled data is scarce. We derive two upper bounds for the robust risk and propose a regularization term for unlabeled data motivated by these two upper bounds. Then, we develop a semi-supervised adversarial training algorithm that combines the proposed regularization term with knowledge distillation using a semi-supervised teacher (i.e., a teacher model trained using a semi-supervised learning algorithm). Our experiments show that our proposed algorithm achieves state-of-the-art performance with significant margins compared to existing algorithms. In particular, compared to supervised learning algorithms, performance of our proposed algorithm is not much worse even when the amount of labeled data is very small. For example, our algorithm with only 8\% labeled data is comparable to supervised adversarial training algorithms that use all labeled data, both in terms of standard and robust accuracies on CIFAR-10.

arXiv:2308.02525공식 코드 · Papers with Code

Can Self-Supervised Representation Learning Methods Withstand Distribution Shifts and Corruptions?

Self-supervised learning in computer vision aims to leverage the inherent structure and relationships within data to learn meaningful representations without explicit human annotation, enabling a holistic understanding of visual scenes. Robustness in vision machine learning ensures reliable and consistent performance, enhancing generalization, adaptability, and resistance to noise, variations, and adversarial attacks. Self-supervised paradigms, namely contrastive learning, knowledge distillation, mutual information maximization, and clustering, have been considered to have shown advances in invariant learning representations. This work investigates the robustness of learned representations of self-supervised learning approaches focusing on distribution shifts and image corruptions in computer vision. Detailed experiments have been conducted to study the robustness of self-supervised learning methods on distribution shifts and image corruptions. The empirical analysis demonstrates a clear relationship between the performance of learned representations within self-supervised paradigms and the severity of distribution shifts and corruptions. Notably, higher levels of shifts and corruptions are found to significantly diminish the robustness of the learned representations. These findings highlight the critical impact of distribution shifts and image corruptions on the performance and resilience of self-supervised learning methods, emphasizing the need for effective strategies to mitigate their adverse effects. The study strongly advocates for future research in the field of self-supervised representation learning to prioritize the key aspects of safety and robustness in order to ensure practical applicability. The source code and results are available on GitHub.

arXiv:2308.02790공식 코드 · Papers with Code

Few-shot Class-Incremental Semantic Segmentation via Pseudo-Labeling and Knowledge Distillation

We address the problem of learning new classes for semantic segmentation models from few examples, which is challenging because of the following two reasons. Firstly, it is difficult to learn from limited novel data to capture the underlying class distribution. Secondly, it is challenging to retain knowledge for existing classes and to avoid catastrophic forgetting. For learning from limited data, we propose a pseudo-labeling strategy to augment the few-shot training annotations in order to learn novel classes more effectively. Given only one or a few images labeled with the novel classes and a much larger set of unlabeled images, we transfer the knowledge from labeled images to unlabeled images with a coarse-to-fine pseudo-labeling approach in two steps. Specifically, we first match each labeled image to its nearest neighbors in the unlabeled image set at the scene level, in order to obtain images with a similar scene layout. This is followed by obtaining pseudo-labels within this neighborhood by applying classifiers learned on the few-shot annotations. In addition, we use knowledge distillation on both labeled and unlabeled data to retain knowledge on existing classes. We integrate the above steps into a single convolutional neural network with a unified learning objective. Extensive experiments on the Cityscapes and KITTI datasets validate the efficacy of the proposed approach in the self-driving domain. Code is available from https://github.com/ChasonJiang/FSCILSS.

arXiv:2308.02770공식 코드 · Papers with Code

One-stage Low-resolution Text Recognition with High-resolution Knowledge Transfer

Recognizing characters from low-resolution (LR) text images poses a significant challenge due to the information deficiency as well as the noise and blur in low-quality images. Current solutions for low-resolution text recognition (LTR) typically rely on a two-stage pipeline that involves super-resolution as the first stage followed by the second-stage recognition. Although this pipeline is straightforward and intuitive, it has to use an additional super-resolution network, which causes inefficiencies during training and testing. Moreover, the recognition accuracy of the second stage heavily depends on the reconstruction quality of the first stage, causing ineffectiveness. In this work, we attempt to address these challenges from a novel perspective: adapting the recognizer to low-resolution inputs by transferring the knowledge from the high-resolution. Guided by this idea, we propose an efficient and effective knowledge distillation framework to achieve multi-level knowledge transfer. Specifically, the visual focus loss is proposed to extract the character position knowledge with resolution gap reduction and character region focus, the semantic contrastive loss is employed to exploit the contextual semantic knowledge with contrastive learning, and the soft logits loss facilitates both local word-level and global sequence-level learning from the soft teacher label. Extensive experiments show that the proposed one-stage pipeline significantly outperforms super-resolution based two-stage frameworks in terms of effectiveness and efficiency, accompanied by favorable robustness. Code is available at https://github.com/csguoh/KD-LTR.

arXiv:2308.03725공식 코드 · Papers with Code

Efficient Temporal Sentence Grounding in Videos with Multi-Teacher Knowledge Distillation

Temporal Sentence Grounding in Videos (TSGV) aims to detect the event timestamps described by the natural language query from untrimmed videos. This paper discusses the challenge of achieving efficient computation in TSGV models while maintaining high performance. Most existing approaches exquisitely design complex architectures to improve accuracy with extra layers and loss, suffering from inefficiency and heaviness. Although some works have noticed that, they only make an issue of feature fusion layers, which can hardly enjoy the highspeed merit in the whole clunky network. To tackle this problem, we propose a novel efficient multi-teacher model (EMTM) based on knowledge distillation to transfer diverse knowledge from both heterogeneous and isomorphic networks. Specifically, We first unify different outputs of the heterogeneous models into one single form. Next, a Knowledge Aggregation Unit (KAU) is built to acquire high-quality integrated soft labels from multiple teachers. After that, the KAU module leverages the multi-scale video and global query information to adaptively determine the weights of different teachers. A Shared Encoder strategy is then proposed to solve the problem that the student shallow layers hardly benefit from teachers, in which an isomorphic teacher is collaboratively trained with the student to align their hidden states. Extensive experimental results on three popular TSGV benchmarks demonstrate that our method is both effective and efficient without bells and whistles.

arXiv:2308.02019공식 코드 · Papers with Code

Baby Llama: knowledge distillation from an ensemble of teachers trained on a small dataset with no performance penalty

We present our submission to the BabyLM challenge, whose goal was to improve the sample efficiency of language models. We trained an ensemble consisting of a GPT-2 and small LLaMA models on the developmentally-plausible, 10M-word BabyLM dataset, then distilled it into a small, 58M-parameter LLaMA model, which exceeds in performance both of its teachers as well as a similar model trained without distillation. This suggests that distillation can not only retain the full performance of the teacher model when the latter is trained on a sufficiently small dataset; it can exceed it, and lead to significantly better performance than direct training.

arXiv:2308.02117공식 코드 · Papers with Code

VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs

GNN-to-MLP distillation aims to utilize knowledge distillation (KD) to learn computationally-efficient multi-layer perceptron (student MLP) on graph data by mimicking the output representations of teacher GNN. Existing methods mainly make the MLP to mimic the GNN predictions over a few class labels. However, the class space may not be expressive enough for covering numerous diverse local graph structures, thus limiting the performance of knowledge transfer from GNN to MLP. To address this issue, we propose to learn a new powerful graph representation space by directly labeling nodes' diverse local structures for GNN-to-MLP distillation. Specifically, we propose a variant of VQ-VAE to learn a structure-aware tokenizer on graph data that can encode each node's local substructure as a discrete code. The discrete codes constitute a codebook as a new graph representation space that is able to identify different local graph structures of nodes with the corresponding code indices. Then, based on the learned codebook, we propose a new distillation target, namely soft code assignments, to directly transfer the structural knowledge of each node from GNN to MLP. The resulting framework VQGraph achieves new state-of-the-art performance on GNN-to-MLP distillation in both transductive and inductive settings across seven graph datasets. We show that VQGraph with better performance infers faster than GNNs by 828x, and also achieves accuracy improvement over GNNs and stand-alone MLPs by 3.90% and 28.05% on average, respectively. Code: https://github.com/YangLing0818/VQGraph.

arXiv:2308.01126공식 코드 · Papers with Code

Beyond Generic: Enhancing Image Captioning with Real-World Knowledge using Vision-Language Pre-Training Model

Current captioning approaches tend to generate correct but "generic" descriptions that lack real-world knowledge, e.g., named entities and contextual information. Considering that Vision-Language Pre-Training (VLP) models master massive such knowledge from large-scale web-harvested data, it is promising to utilize the generalizability of VLP models to incorporate knowledge into image descriptions. However, using VLP models faces challenges: zero-shot inference suffers from knowledge hallucination that leads to low-quality descriptions, but the generic bias in downstream task fine-tuning hinders the VLP model from expressing knowledge. To address these concerns, we propose a simple yet effective method called Knowledge-guided Replay (K-Replay), which enables the retention of pre-training knowledge during fine-tuning. Our approach consists of two parts: (1) a knowledge prediction task on automatically collected replay exemplars to continuously awaken the VLP model's memory about knowledge, thus preventing the model from collapsing into the generic pattern; (2) a knowledge distillation constraint to improve the faithfulness of generated descriptions hence alleviating the knowledge hallucination. To evaluate knowledge-enhanced descriptions, we construct a novel captioning benchmark KnowCap, containing knowledge of landmarks, famous brands, special foods and movie characters. Experimental results show that our approach effectively incorporates knowledge into descriptions, outperforming strong VLP baseline by 20.9 points (78.7->99.6) in CIDEr score and 20.5 percentage points (34.0%->54.5%) in knowledge recognition accuracy. Our code and data is available at https://github.com/njucckevin/KnowCap.

arXiv:2308.01097공식 코드 · Papers with Code

Spatio-Temporal Branching for Motion Prediction using Motion Increments

Human motion prediction (HMP) has emerged as a popular research topic due to its diverse applications, but it remains a challenging task due to the stochastic and aperiodic nature of future poses. Traditional methods rely on hand-crafted features and machine learning techniques, which often struggle to model the complex dynamics of human motion. Recent deep learning-based methods have achieved success by learning spatio-temporal representations of motion, but these models often overlook the reliability of motion data. Additionally, the temporal and spatial dependencies of skeleton nodes are distinct. The temporal relationship captures motion information over time, while the spatial relationship describes body structure and the relationships between different nodes. In this paper, we propose a novel spatio-temporal branching network using incremental information for HMP, which decouples the learning of temporal-domain and spatial-domain features, extracts more motion information, and achieves complementary cross-domain knowledge learning through knowledge distillation. Our approach effectively reduces noise interference and provides more expressive information for characterizing motion by separately extracting temporal and spatial features. We evaluate our approach on standard HMP benchmarks and outperform state-of-the-art methods in terms of prediction accuracy.

arXiv:2308.00301공식 코드 · Papers with Code

Online Prototype Learning for Online Continual Learning

Online continual learning (CL) studies the problem of learning continuously from a single-pass data stream while adapting to new data and mitigating catastrophic forgetting. Recently, by storing a small subset of old data, replay-based methods have shown promising performance. Unlike previous methods that focus on sample storage or knowledge distillation against catastrophic forgetting, this paper aims to understand why the online learning models fail to generalize well from a new perspective of shortcut learning. We identify shortcut learning as the key limiting factor for online CL, where the learned features may be biased, not generalizable to new tasks, and may have an adverse impact on knowledge distillation. To tackle this issue, we present the online prototype learning (OnPro) framework for online CL. First, we propose online prototype equilibrium to learn representative features against shortcut learning and discriminative features to avoid class confusion, ultimately achieving an equilibrium status that separates all seen classes well while learning new classes. Second, with the feedback of online prototypes, we devise a novel adaptive prototypical feedback mechanism to sense the classes that are easily misclassified and then enhance their boundaries. Extensive experimental results on widely-used benchmark datasets demonstrate the superior performance of OnPro over the state-of-the-art baseline methods. Source code is available at https://github.com/weilllllls/OnPro.

arXiv:2308.00520공식 코드 · Papers with Code

NormKD: Normalized Logits for Knowledge Distillation

Logit based knowledge distillation gets less attention in recent years since feature based methods perform better in most cases. Nevertheless, we find it still has untapped potential when we re-investigate the temperature, which is a crucial hyper-parameter to soften the logit outputs. For most of the previous works, it was set as a fixed value for the entire distillation procedure. However, as the logits from different samples are distributed quite variously, it is not feasible to soften all of them to an equal degree by just a single temperature, which may make the previous work transfer the knowledge of each sample inadequately. In this paper, we restudy the hyper-parameter temperature and figure out its incapability to distill the knowledge from each sample sufficiently when it is a single value. To address this issue, we propose Normalized Knowledge Distillation (NormKD), with the purpose of customizing the temperature for each sample according to the characteristic of the sample's logit distribution. Compared to the vanilla KD, NormKD barely has extra computation or storage cost but performs significantly better on CIRAR-100 and ImageNet for image classification. Furthermore, NormKD can be easily applied to the other logit based methods and achieve better performance which can be closer to or even better than the feature based method.

arXiv:2307.15880공식 코드 · Papers with Code

Effective Whole-body Pose Estimation with Two-stages Distillation

Whole-body pose estimation localizes the human body, hand, face, and foot keypoints in an image. This task is challenging due to multi-scale body parts, fine-grained localization for low-resolution regions, and data scarcity. Meanwhile, applying a highly efficient and accurate pose estimator to widely human-centric understanding and generation tasks is urgent. In this work, we present a two-stage pose \textbf{D}istillation for \textbf{W}hole-body \textbf{P}ose estimators, named \textbf{DWPose}, to improve their effectiveness and efficiency. The first-stage distillation designs a weight-decay strategy while utilizing a teacher's intermediate feature and final logits with both visible and invisible keypoints to supervise the student from scratch. The second stage distills the student model itself to further improve performance. Different from the previous self-knowledge distillation, this stage finetunes the student's head with only 20% training time as a plug-and-play training strategy. For data limitations, we explore the UBody dataset that contains diverse facial expressions and hand gestures for real-life applications. Comprehensive experiments show the superiority of our proposed simple yet effective methods. We achieve new state-of-the-art performance on COCO-WholeBody, significantly boosting the whole-body AP of RTMPose-l from 64.8% to 66.5%, even surpassing RTMPose-x teacher with 65.3% AP. We release a series of models with different sizes, from tiny to large, for satisfying various downstream tasks. Our codes and models are available at https://github.com/IDEA-Research/DWPose.

arXiv:2307.16419공식 코드 · Papers with Code

Subspace Distillation for Continual Learning

An ultimate objective in continual learning is to preserve knowledge learned in preceding tasks while learning new tasks. To mitigate forgetting prior knowledge, we propose a novel knowledge distillation technique that takes into the account the manifold structure of the latent/output space of a neural network in learning novel tasks. To achieve this, we propose to approximate the data manifold up-to its first order, hence benefiting from linear subspaces to model the structure and maintain the knowledge of a neural network while learning novel concepts. We demonstrate that the modeling with subspaces provides several intriguing properties, including robustness to noise and therefore effective for mitigating Catastrophic Forgetting in continual learning. We also discuss and show how our proposed method can be adopted to address both classification and segmentation problems. Empirically, we observe that our proposed method outperforms various continual learning methods on several challenging datasets including Pascal VOC, and Tiny-Imagenet. Furthermore, we show how the proposed method can be seamlessly combined with existing learning approaches to improve their performances. The codes of this article will be available at https://github.com/csiro-robotics/SDCL.