The excellent performance of deep neural networks is usually accompanied by a large number of parameters and computations, which have limited their usage on the resource-limited edge devices. To address this issue, abundant methods such as pruning, quantization and knowledge distillation have been proposed to compress neural networks and achieved significant breakthroughs. However, most of these compression methods focus on the architecture or the training method of neural networks but ignore the influence from data augmentation. In this paper, we revisit the usage of data augmentation in model compression and give a comprehensive study on the relation between model sizes and their optimal data augmentation policy. To sum up, we mainly have the following three observations: (A) Models in different sizes prefer data augmentation with different magnitudes. Hence, in iterative pruning, data augmentation with varying magnitudes leads to better performance than data augmentation with a consistent magnitude. (B) Data augmentation with a high magnitude may significantly improve the performance of large models but harm the performance of small models. Fortunately, small models can still benefit from strong data augmentations by firstly learning them with "additional parameters" and then discard these "additional parameters" during inference. (C) The prediction of a pre-trained large model can be utilized to measure the difficulty of data augmentation. Thus it can be utilized as a criterion to design better data augmentation policies. We hope this paper may promote more research on the usage of data augmentation in model compression.
ETH Zürich（苏黎世联邦理工学院）；MPI-IS（马克斯·普朗克智能系统研究所）；UESTC（电子科技大学）；USYD（悉尼大学）；KU Leuven（鲁汶大学）
Channel (or 3D filter) pruning serves as an effective way to accelerate the inference of neural networks. There has been a flurry of algorithms that try to solve this practical problem, each being claimed effective in some ways. Yet, a benchmark to compare those algorithms directly is lacking, mainly due to the complexity of the algorithms and some custom settings such as the particular network configuration or training procedure. A fair benchmark is important for the further development of channel pruning. Meanwhile, recent investigations reveal that the channel configurations discovered by p…
Facebook AI Research（Facebook人工智能研究院）；Univ Rennes（雷恩大学）；Inria（法国国家信息与自动化研究所）；CNRS（法国国家科学研究中心）；IRISA（IRISA研究所）
In this paper, we address the problem of reducing the memory footprint of convolutional network architectures. We introduce a vector quantization method that aims at preserving the quality of the reconstruction of the network outputs rather than its weights. The principle of our approach is that it minimizes the loss reconstruction error for in-domain inputs. Our method only requires a set of unlabelled data at quantization time and allows for efficient inference on CPU by using byte-aligned codebooks to store the compressed weights. We validate our approach by quantizing a high performing Res…
The University of Sydney（悉尼大学）；Swiss Federal Institute of Technology in Zürich（苏黎世联邦理工学院）；University of Electronic Science and Technology of China（电子科技大学）
In this work, we revisit the semi-supervised learning (SSL) problem from a new perspective of explicitly reducing empirical distribution mismatch between labeled and unlabeled samples. Benefited from this new perspective, we first propose a new deep semi-supervised learning framework called Semi-supervised Learning by Empirical Distribution Alignment (SLEDA), in which existing technologies from the domain adaptation community can be readily used to address the semi-supervised learning problem through reducing the empirical distribution distance between labeled and unlabeled data. Based on this…
Recently, a series of algorithms have been explored for GAN compression, which aims to reduce tremendous computational overhead and memory usages when deploying GANs on resource-constrained edge devices. However, most of the existing GAN compression work only focuses on how to compress the generator, while fails to take the discriminator into account. In this work, we revisit the role of discriminator in GAN compression and design a novel generator-discriminator cooperative compression scheme for GAN compression, termed GCC. Within GCC, a selective activation discriminator automatically select…
Boosting Semi-Supervised 2D Human Pose Estimation by Revisiting Data Augmentation and Consistency Training
School of Data Science, The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳）数据科学学院）；Chongqing Academy of Science and Technology（重庆科学技术研究院）；Department of Computer Science and Engineering, Shanghai Jiao Tong University（上海交通大学计算机科学与工程系）
We focus on better understanding the critical factors of augmentation-invariant representation learning. We revisit MoCo v2 and BYOL and try to prove the authenticity of the following assumption: different frameworks bring about representations of different characteristics even with the same pretext task. We establish the first benchmark for fair comparisons between MoCo v2 and BYOL, and observe: (i) sophisticated model configurations enable better adaptation to pre-training dataset; (ii) mismatched optimization strategies of pre-training and fine-tuning hinder model from achieving competitive…
UCSF（加州大学旧金山分校）；SFVAMC（旧金山退伍军人事务医疗中心）；UC Berkeley（加州大学伯克利分校）；Helen Diller Family Comprehensive Cancer Center, UCSF（加州大学旧金山分校海伦·迪勒家族综合癌症中心）；Bakar Computational Health Sciences Institute, UCSF（加州大学旧金山分校巴卡尔计算健康科学研究所）
Interpretability methods for image classification assess model trustworthiness by attempting to expose whether the model is systematically biased or attending to the same cues as a human would. Saliency methods for feature attribution dominate the interpretability literature, but these methods do not address semantic concepts such as the textures, colors, or genders of objects within an image. Our proposed Robust Concept Activation Vectors (RCAV) quantifies the effects of semantic concepts on individual model predictions and on model behavior as a whole. RCAV calculates a concept gradient and …
What role do augmentations play in contrastive learning? Recent work suggests that good augmentations are label-preserving with respect to a specific downstream task. We complicate this picture by showing that label-destroying augmentations can be useful in the foundation model setting, where the goal is to learn diverse, general-purpose representations for multiple downstream tasks. We perform contrastive learning experiments on a range of image and audio datasets with multiple downstream tasks (e.g. for digits superimposed on photographs, predicting the class of one vs. the other). We find t…
Revisiting Data Augmentation for Rotational Invariance in Convolutional Neural Networks
Instituto de Investigación en Informática LIDI, Facultad de Informática, Universidad Nacional de La Plata（拉普拉塔国立大学信息学院LIDI信息学研究所）；Universitat Rovira i Virgili（罗维拉-依维尔基里大学）
Subspace clustering is the classical problem of clustering a collection of data samples that approximately lie around several low-dimensional subspaces. The current state-of-the-art approaches for this problem are based on the self-expressive model which represents the samples as linear combination of other samples. However, these approaches require sufficiently well-spread samples for accurate representation which might not be necessarily accessible in many applications. In this paper, we shed light on this commonly neglected issue and argue that data distribution within each subspace plays a…
Revisiting Token Compression for Accelerating ViT-based Sparse Multi-View 3D Object Detectors
PCA Lab, School of Computer Science and Engineering, Nanjing University of Science and Technology（计算机科学与工程学院，南京理工大学PCA实验室）；PCA Lab, School of Intelligence Science and Technology, Nanjing University（智能科学与技术学院，南京大学PCA实验室）
Max Planck Institute for Informatics（马克斯·普朗克信息学研究所）
Consistency regularization is one of the most widely-used techniques for semi-supervised learning (SSL). Generally, the aim is to train a model that is invariant to various data augmentations. In this paper, we revisit this idea and find that enforcing invariance by decreasing distances between features from differently augmented images leads to improved performance. However, encouraging equivariance instead, by increasing the feature distance, further improves performance. To this end, we propose an improved consistency regularization framework by a simple yet effective technique, FeatDistLos…
Revisiting Multimodal KV Cache Compression: A Frequency-Domain-Guided Outlier-KV-Aware Approach
College of Future Information Technology, Fudan University（未来信息科技学院，复旦大学）；Shanghai Innovation Institute（上海创新研究院）；The Chinese University of Hong Kong（香港中文大学）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；Zhangjiang Laboratory（张江实验室）
Mila, Université de Montréal（蒙特利尔大学 Mila 研究院）；HCI/IWR, Heidelberg University（海德堡大学 HCI/IWR）；University of Toronto（多伦多大学）；The Alan Turing Institute, University of Oxford（牛津大学艾伦·图灵研究所）
Deep Metric Learning (DML) is arguably one of the most influential lines of research for learning visual similarities with many proposed approaches every year. Although the field benefits from the rapid progress, the divergence in training protocols, architectures, and parameter choices make an unbiased comparison difficult. To provide a consistent reference point, we revisit the most widely used DML objective functions and conduct a study of the crucial parameter choices as well as the commonly neglected mini-batch sampling process. Under consistent comparison, DML objectives show much higher…
Upgrad Education Pvt. Ltd.（Upgrad教育私人有限公司）；Salesforce（赛富时）；Liverpool John Moore University（利物浦约翰摩尔大学）
Facial landmark detection is a widely researched field of deep learning as this has a wide range of applications in many fields. These key points are distinguishing characteristic points on the face, such as the eyes center, the eye's inner and outer corners, the mouth center, and the nose tip from which human emotions and intent can be explained. The focus of our work has been evaluating transfer learning models such as MobileNetV2 and NasNetMobile, including custom CNN architectures. The objective of the research has been to develop efficient deep learning models in terms of model size, para…
Department of Computer Science, Texas State University, USA（德克萨斯州立大学计算机科学系，美国（注：原机构名中Department of Computer Science属于系级别，按要求提取到大学级别，即Texas State University，美国德克萨斯州立大学））
Nowadays 360 video analysis has become a significant research topic in the field since the appearance of high-quality and low-cost 360 wearable devices. In this paper, we propose a novel LiteFlowNet360 architecture for 360 videos optical flow estimation. We design LiteFlowNet360 as a domain adaptation framework from perspective video domain to 360 video domain. We adapt it from simple kernel transformation techniques inspired by Kernel Transformer Network (KTN) to cope with inherent distortion in 360 videos caused by the sphere-to-plane projection. First, we apply an incremental transformation…
Diffusion probabilistic models have recently achieved remarkable success in generating high quality image and video data. In this work, we build on this class of generative models and introduce a method for lossy compression of high resolution images. The resulting codec, which we call DIffuson-based Residual Augmentation Codec (DIRAC), is the first neural codec to allow smooth traversal of the rate-distortion-perception tradeoff at test time, while obtaining competitive performance with GAN-based methods in perceptual quality. Furthermore, while sampling from diffusion probabilistic models is…
A large number of coils are able to provide enhanced signal-to-noise ratio and improve imaging performance in parallel imaging. Nevertheless, the increasing growth of coil number simultaneously aggravates the drawbacks of data storage and reconstruction speed, especially in some iterative reconstructions. Coil compression addresses these issues by generating fewer virtual coils. In this work, a novel variable augmentation network for invertible coil compression termed VAN-ICC is presented. It utilizes inherent reversibility of normalizing flow-based models for high-precision compression and in…
We present a deep person re-identification approach that combines semantically selective, deep data augmentation with clustering-based network compression to generate high performance, light and fast inference networks. In particular, we propose to augment limited training data via sampling from a deep convolutional generative adversarial network (DCGAN), whose discriminator is constrained by a semantic classifier to explicitly control the domain specificity of the generation process. Thereby, we encode information in the classifier network which can be utilized to steer adversarial synthesis,…
Processing point cloud data is an important component of many real-world systems. As such, a wide variety of point-based approaches have been proposed, reporting steady benchmark improvements over time. We study the key ingredients of this progress and uncover two critical results. First, we find that auxiliary factors like different evaluation schemes, data augmentation strategies, and loss functions, which are independent of the model architecture, make a large difference in performance. The differences are large enough that they obscure the effect of architecture. When these factors are con…
The Ohio State University（俄亥俄州立大学）；Nanjing University（南京大学）；Cornell University（康奈尔大学）
Recent years have witnessed an abundance of new publications and approaches on meta-learning. This community-wide enthusiasm has sparked great insights but has also created a plethora of seemingly different frameworks, which can be hard to compare and evaluate. In this paper, we aim to provide a principled, unifying framework by revisiting and strengthening the connection between meta-learning and traditional supervised learning. By treating pairs of task-specific data sets and target models as (feature, label) samples, we can reduce many meta-learning algorithms to instances of supervised lea…
Revisiting Energy-Based Model for Out-of-Distribution Detection
School of Computer Engineering and Science, Shanghai University（上海大学计算机工程与科学学院）；School of Computer Science, Fudan University（复旦大学计算机科学学院）；School of Data Science, Fudan University（复旦大学数据科学学院）；Department of Computer Science, Hong Kong Baptist University（香港浸会大学计算机科学系）
Zhejiang Gongshang University（浙江工商大学）；Renmin University of China（中国人民大学）
Predicting the relevance between two given videos with respect to their visual content is a key component for content-based video recommendation and retrieval. Thanks to the increasing availability of pre-trained image and video convolutional neural network models, deep visual features are widely used for video content representation. However, as how two videos are relevant is task-dependent, such off-the-shelf features are not always optimal for all tasks. Moreover, due to varied concerns including copyright, privacy and security, one might have access to only pre-computed video features rath…
Anomaly Detection (AD) on medical images enables a model to recognize any type of anomaly pattern without lesion-specific supervised learning. Data augmentation based methods construct pseudo-healthy images by "pasting" fake lesions on real healthy ones, and a network is trained to predict healthy images in a supervised manner. The lesion can be found by difference between the unhealthy input and pseudo-healthy output. However, using only manually designed fake lesions fail to approximate to irregular real lesions, hence limiting the model generalization. We assume by exploring the intrinsic d…
Data augmentation is an effective way to improve the performance of deep networks. Unfortunately, current methods are mostly developed for high-level vision tasks (e.g., classification) and few are studied for low-level vision tasks (e.g., image restoration). In this paper, we provide a comprehensive analysis of the existing augmentation methods applied to the super-resolution task. We find that the methods discarding or manipulating the pixels or features too much hamper the image restoration, where the spatial relationship is very important. Based on our analyses, we propose CutBlur that cut…
École Polytechnique Fédérale de Lausanne (EPFL)（洛桑联邦理工学院）
Deep convolutional neural networks have achieved exceptional results on multiple detection and recognition tasks. However, the performance of such detectors are often evaluated in public benchmarks under constrained and non-realistic situations. The impact of conventional distortions and processing operations found in imaging workflows such as compression, noise, and enhancement are not sufficiently studied. Currently, only a few researches have been done to improve the detector robustness to unseen perturbations. This paper proposes a more effective data augmentation scheme based on real-worl…
Multimedia Systems Department（多媒体系统系）；Gdańsk University of Technology（格但斯克工业大学）；Electronics, Telecommunications and Informatics Faculty（电子、电信与信息学院）；Polish National Centre for Research and Development (NCBR)（波兰国家研究与发展中心）
Model compression techniques allow to significantly reduce the computational cost associated with data processing by deep neural networks with only a minor decrease in average accuracy. Simultaneously, reducing the model size may have a large effect on noisy cases or objects belonging to less frequent classes. It is a crucial problem from the perspective of the models' safety, especially for object detection in the autonomous driving setting, which is considered in this work. It was shown in the paper that the sensitivity of compressed models to different distortion types is nuanced, and some …
The recent advancement in deep Convolutional Neural Network (CNN) has brought insight into the automation of X-ray security screening for aviation security and beyond. Here, we explore the viability of two recent end-to-end object detection CNN architectures, Cascade R-CNN and FreeAnchor, for prohibited item detection by balancing processing time and the impact of image data compression from an operational viewpoint. Overall, we achieve maximal detection performance using a FreeAnchor architecture with a ResNet50 backbone, obtaining mean Average Precision (mAP) of 87.7 and 85.8 for using the O…
