论文检索

Deep Learning/Generative Models and Autoencoders

Martin Gonzalez, Nelson Fernandez Pinto, Thuy Tran, elies Gherbi, Hatem Hajri, Nader Masmoudi

A potent class of generative models known as Diffusion Probabilistic Models(DPMs) has become prominent. A forward diffusion process adds gradually noiseto data, while a model learns to gradually denoise. Sampling from pre-trainedDPMs is obtained by solving differential equations (DE) defined by the learntmodel, a process which has shown to be prohibitively slow. Numerous efforts onspeeding-up this process have consisted on crafting powerful ODE solvers.Despite being quick, such solvers do not usually reach the optimal qualityachieved by available slow SDE solvers. Our goal is to propose SDE solvers thatreach optimal quality without requiring several hundreds or thousands of NFEsto achieve that goal. We propose Stochastic Explicit ExponentialDerivative-free Solvers (SEEDS), improving and generalizing ExponentialIntegrator approaches to the stochastic case on several frameworks. After carefully analyzing the formulation of exactsolutions of diffusion SDEs, we craft SEEDS to analytically compute the linearpart of such solutions. Inspired by the Exponential Time-Differencing method,SEEDS use a novel treatment of the stochastic components of solutions,enabling the analytical computation of their variance, and contains high-orderterms allowing to reach optimal quality sampling $\sim3$-$5\times$ faster than previousSDE methods. We validate our approach on several image generation benchmarks,showing that SEEDS outperform or are competitive with previous SDE solvers.Contrary to the latter, SEEDS are derivative and training free, and we fullyprove strong convergence guarantees for them.

Fabian Duffhauss, Ngo Anh Vien, Hanna Ziesche, Gerhard Neumann

Sensor fusion can significantly improve the performance of many computer vision tasks. However, traditional fusion approaches are either not data-driven and cannot exploit prior knowledge nor find regularities in a given dataset or they are restricted to a single application. We overcome this shortcoming by presenting a novel deep hierarchical variational autoencoder called FusionVAE that can serve as a basis for many fusion tasks. Our approach is able to generate diverse image samples that are conditioned on multiple noisy, occluded, or only partially visible input images. We derive and optimize a variational lower bound for the conditional log-likelihood of FusionVAE. In order to assess the fusion capabilities of our model thoroughly, we created three novel datasets for image fusion based on popular computer vision datasets. In our experiments, we show that FusionVAE learns a representation of aggregated information that is relevant to fusion tasks. The results demonstrate that our approach outperforms traditional methods significantly. Furthermore, we present the advantages and disadvantages of different design choices.

Marah Halawa, Olaf Hellwich, Pia Bideau

Trajectory prediction is an essential task for successful human robot interaction, such as in autonomous driving. In this work, we address the problem of predicting future pedestrian trajectories in a first person view setting with a moving camera. To that end, we propose a novel action-based contrastive learning loss, that utilizes pedestrian action information to improve the learned trajectory embeddings. The fundamental idea behind this new loss is that trajectories of pedestrians performing the same action should be closer to each other in the feature space than the trajectories of pedestrians with significantly different actions. In other words, we argue that behavioral information about pedestrian action influences their future trajectory. Furthermore, we introduce a novel sampling strategy for trajectories that is able to effectively increase negative and positive contrastive samples. Additional synthetic trajectory samples are generated using a trained Conditional Variational Autoencoder (CVAE), which is at the core of several models developed for trajectory prediction. Results show that our proposed contrastive framework employs contextual information about pedestrian behavior, i.e. action, effectively, and it learns a better trajectory representation. Thus, integrating the proposed contrastive framework within a trajectory prediction model improves its results and outperforms state-of-the-art methods on three trajectory prediction benchmarks.

Roman Bachmann, David Mizrahi, Andrei Atanov, Amir Zamir

We propose a pre-training strategy called Multi-modal Multi-task Masked Autoencoders (MultiMAE). It differs from standard Masked Autoencoding in two key aspects: I) it can ‘optionally’ accept additional modalities of information in the input besides the RGB image (hence “multi-modal""""), and II) its training objective accordingly includes predicting multiple outputs besides the RGB image (hence “multi-task""""). We make use of masking (across image patches and input modalities) to make training MultiMAE tractable as well as to ensure cross-modality predictive coding is indeed learned by the network. We show this pre-training strategy leads to a flexible, simple, and efficient framework with improved transfer results to downstream tasks. In particular, the same exact pre-trained network can be flexibly used when additional information besides RGB images is available or when no information other than RGB is available - in all configurations yielding competitive to or significantly better results than the baselines. To avoid needing training datasets with multiple modalities and tasks, we train MultiMAE ‘entirely using pseudo-labeling’, which makes the framework widely applicable to any RGB dataset. The experiments are performed on multiple transfer tasks (image classification, semantic segmentation, depth estimation) on different datasets (ImageNet, ADE20K, Taskonomy, Hypersim, NYUv2), and the results show an intriguingly impressive capability by the network in cross-modal/task predictive coding and transfer. Code, pre-trained models, and interactive visualizations are available at https://multimae.epfl.ch.

Guy Erez, Ron Shapira Weber, Oren Freifeld

In video analysis, background models have many applications such as background/foreground separation, change detection, anomaly detection, tracking, and more. However, while learning such a model in a video captured by a static camera is a fairly-solved task, in the case of a Moving-camera Background Model (MCBM), the success has been far more modest due to algorithmic and scalability challenges that arise due to the camera motion. Thus, existing MCBMs are limited in their scope and their supported camera-motion types. These hurdles also impeded the employment, in this unsupervised task, of end-to-end solutions based on deep learning (DL). Moreover, existing MCBMs usually model the background either on the domain of a typically-large panoramic image or in an online fashion. Unfortunately, the former creates several problems, including poor scalability, while the latter prevents the recognition and leveraging of cases where the camera revisits previously-seen parts of the scene. This paper proposes a new method, called DeepMCBM, that eliminates all the aforementioned issues and achieves state-of-the-art results. Concretely, first we identify the difficulties associated with joint alignment of video frames in general and in a DL setting in particular. Next, we propose a new strategy for joint alignment that lets us use a spatial transformer net with neither a regularization nor any form of specialized (and non-differentiable) initialization. Coupled with an autoencoder conditioned on unwarped robust central moments (obtained from the joint alignment), this yields an end-to-end regularization-free MCBM that supports a broad range of camera motions and scales gracefully. We demonstrate DeepMCBM’s utility on a variety of videos, including ones beyond the scope of other methods. Our code is available at https://github.com/BGU-CS-VIL/DeepMCBM.

Haiyang Yang, Shixiang Tang, Meilin Chen, Yizhou Wang, Feng Zhu, Lei Bai, Rui Zhao, Wanli Ouyang

Generalizing learned representations across significantly different visual domains is a fundamental yet crucial ability of the human visual system. While recent self-supervised learning methods have achieved good performances with evaluation set on the same domain as the training set, they will have an undesirable performance decrease when tested on a different domain. Therefore, the self-supervised learning from multiple domains task is proposed to learn domain-invariant features that are not only suitable for evaluation on the same domain as the training set, but also can be generalized to unseen domains. In this paper, we propose a Domain-invariant Masked AutoEncoder (DiMAE) for self-supervised learning from multi-domains, which designs a new pretext task, i.e., the cross-domain reconstruction task, to learn domain-invariant features. The core idea is to augment the input image with style noise from different domains and then reconstruct the image from the embedding of the augmented image, regularizing the encoder to learn domain-invariant features. To accomplish the idea, DiMAE contains two critical designs, 1) content-preserved style mix, which adds style information from other domains to input while persevering the content in a parameter-free manner, and 2) multiple domain-specific decoders, which recovers the corresponding domain style of input to the encoded domain-invariant features for reconstruction. Experiments on PACS and DomainNet illustrate that DiMAE achieves considerable gains compared with recent state-of-the-art methods.

Xiaoyi Dong, Jianmin Bao, Ting Zhang, Dongdong Chen, Weiming Zhang, Lu Yuan, Dong Chen, Fang Wen, Nenghai Yu

We propose bootstrapped masked autoencoders (BootMAE), a new approach for vision BERT pretraining. BootMAE improves the original masked autoencoders (MAE) with two core designs: 1) momentum encoder that provides online feature as extra BERT prediction targets; 2) target-aware decoder that tries to reduce the pressure on the encoder to memorize target-specific information in BERT pretraining. The first design is motivated by the observation that using a pretrained MAE to extract the features as the BERT prediction target for masked tokens can achieve better pretraining performance. Therefore, we add a momentum encoder in parallel with the original MAE encoder, which bootstraps the pretraining performance by using its own representation as the BERT prediction target. In the second design, we introduce target-specific information (e.g., pixel values of unmasked patches) from the encoder directly to the decoder to reduce the pressure on the encoder of memorizing the target-specific information. Thus, the encoder focuses on semantic modeling, which is the goal of BERT pretraining, and does not need to waste its capacity in memorizing the information of unmasked tokens related to the prediction target. Through extensive experiments, our BootMAE achieves 84.2% Top-1 accuracy on ImageNet-1K with ViT-B backbone, outperforming MAE by +0.8% under the same pre-training epochs. BootMAE also gets +1.0 mIoU improvements on semantic segmentation on ADE20K and +1.3 box AP, +1.4 mask AP improvement on object detection and segmentation on COCO dataset. Code is released at https://github.com/LightDXY/BootMAE

Yabo Chen, Yuchen Liu, Dongsheng Jiang, Xiaopeng Zhang, Wenrui Dai, Hongkai Xiong, Qi Tian

With the development of generative-based self-supervised learning (SSL) approaches like BeiT and MAE, how to learn good representations by masking random patches of the input image and reconstructing the missing information has grown in concern. However, BeiT and PeCo need a “pre-pretraining” stage to produce discrete codebooks for masked patches representing. MAE does not require a pre-training codebook process, but setting pixels as reconstruction targets may introduce an optimization gap between pre-training and downstream tasks that good reconstruction quality may not always lead to the high descriptive capability for the model. Considering the above issues, in this paper, we propose a simple Self-distillated masked AutoEncoder network, namely SdAE. SdAE consists of a student branch using an encoder-decoder structure to reconstruct the missing information, and a teacher branch producing latent representation of masked tokens. We also analyze how to build good views for the teacher branch to produce latent representation from the perspective of information bottleneck. After that, we propose a multi-fold masking strategy to provide multiple masked views with balanced information for boosting the performance, which can also reduce the computational complexity. Our approach generalizes well: with only 300 epochs pre-training, a vanilla ViT-Base model achieves an 84.1% fine-tuning accuracy on ImageNet-1k classification, 48.6 mIOU on ADE20K segmentation, and 48.9 mAP on COCO detection with only 300 epochs pre-training, which surpasses other methods by a considerable margin. Code is available at https://github.com/AbrahamYabo/SdAE.

Eric Yeats, Frank Liu, David Womble, Hai Li

We present a self-supervised method to disentangle factors of variation in high-dimensional data that does not rely on prior knowledge of the underlying variation profile (e.g., no assumptions on the number or distribution of the individual variables to be extracted). In this method which we call NashAE, high-dimensional feature disentanglement is accomplished in the low-dimensional latent space of a standard autoencoder (AE) by promoting the discrepancy between each encoding element and information of the element recovered from all other encoding elements. Disentanglement is promoted efficiently by framing this as a minmax game between the AE and an ensemble of regression networks which each provide an estimate of an element conditioned on an observation of all other elements. We quantitatively compare our approach with leading disentanglement methods using existing disentanglement metrics. Furthermore, we show that NashAE has increased reliability and increased capacity to capture salient data characteristics in the learned latent representation.

Yannick Strümpler, Janis Postels, Ren Yang, Luc Van Gool, Federico Tombari

Implicit Neural Representations (INRs) gained attention as a novel and effective representation for various data types. Recently, prior work applied INRs to image compressing. Such compression algorithms are promising candidates as a general purpose approach for any coordinate-based data modality. However, in order to live up to this promise current INR-based compression algorithms need to improve their rate-distortion performance by a large margin. This work progresses on this problem. First, we propose meta learned initializations for INR-based compression which improves rate-distortion performance. As a side effect it also leads to faster convergence speed. Secondly, we introduce a simple yet highly effective change to the network architecture compared to prior work on INR-based compression. Namely, we combine SIREN networks with positional encodings which improves rate distortion performance. Our contributions to source compression with INRs vastly outperform prior work. We show that our INR-based compression algorithm, meta-learning combined with SIREN and positional encodings, outperforms JPEG2000 and Rate-Distortion Autoencoders on Kodak with 2x reduced dimensionality for the first time and closes the gap on full resolution images. To underline the generality of INR-based source compression, we further perform experiments on 3D shape compression where our method greatly outperforms Draco - a traditional compression algorithm.

Junbo Li, Huan Zhang, Cihang Xie

Patch attack, which introduces a perceptible but localized change to the input image, has gained significant momentum in recent years. In this paper, we propose a unified framework to analyze certified patch defense tasks (including both certified detection and certified recovery) using the recently emerged vision transformer. In addition to the existing patch defense setting where only one patch is considered, we provide the very first study on developing certified detection against the \emph{dual patch attack}, in which the attacker is allowed to adversarially manipulate pixels in two different regions. Benefiting from the recent progress in self-supervised vision transformers (\ie, masked autoencoder), our method achieves state-of-the-art performance in both certified detection and certified recovery of adversarial patches. For certified detection, we improve the performance by up to $\app16\%$ on ImageNet without additional training for a single adversarial patch, and for the first time, can also tackle the more challenging dual patch setting. Our method largely \emph{closes the gap} between detection-based certified robustness and clean image accuracy. For certified recovery, our approach improves certified accuracy by $\app2\%$ on ImageNet across all attack sizes, attaining the new state-of-the-art performance.

Fei Ye, Adrian G. Bors

Due to their inference, data representation and reconstruction properties, Variational Autoencoders (VAE) have been successfully used in continual learning classification tasks. However, their ability to generate images with specifications corresponding to the classes and databases learned during Continual Learning (CL) is not well understood and catastrophic forgetting remains a significant challenge. In this paper, we firstly analyze the forgetting behaviour of VAEs by developing a new theoretical framework that formulates CL as a dynamic optimal transport problem. This framework proves approximate bounds to the data likelihood without requiring the task information and explains how the prior knowledge is lost during the training process. We then propose a novel memory buffering approach, namely the Online Cooperative Memorization (OCM) framework, which consists of a Short-Term Memory (STM) that continually stores recent samples to provide future information for the model, and a Long-Term Memory (LTM) aiming to preserve a wide diversity of samples. The proposed OCM transfers certain samples from STM to LTM according to the information diversity selection criterion without requiring any supervised signals. The OCM framework is then combined with a dynamic VAE expansion mixture network for further enhancing its performance.

Mathis Petrovich, Michael J. Black, Gül Varol

We address the problem of generating diverse 3D human motions from textual descriptions. This challenging task requires joint modeling of both modalities: understanding and extracting useful human-centric information from the text, and then generating plausible and realistic sequences of human poses. In contrast to most previous work which focuses on generating a single, deterministic, motion from a textual description, we design a variational approach that can produce multiple diverse human motions. We propose TEMOS, a text-conditioned generative model leveraging variational autoencoder (VAE) training with human motion data, in combination with a text encoder that produces distribution parameters compatible with the VAE latent space. We show the TEMOS framework can produce both skeleton-based animations as in prior work, as well more expressive SMPL body motions. We evaluate our approach on the KIT Motion-Language benchmark and, despite being relatively straightforward, demonstrate significant improvements over the state of the art. Code and models are available on our webpage.

Dooseop Choi, KyoungWook Min

Variational autoencoder (VAE) has widely been utilized for modeling data distributions because it is theoretically elegant, easy to train, and has nice manifold representations. However, when applied to image reconstruction and synthesis tasks, VAE shows the limitation that the generated sample tends to be blurry. We observe that a similar problem, in which the generated trajectory is located between adjacent lanes, often arises in VAE-based trajectory forecasting models. To mitigate this problem, we introduce a hierarchical latent structure into the VAE-based forecasting model. Based on the assumption that the trajectory distribution can be approximated as a mixture of simple distributions (or modes), the low-level latent variable is employed to model each mode of the mixture and the high-level latent variable is employed to represent the weights for the modes. To model each mode accurately, we condition the low-level latent variable using two lane-level context vectors computed in novel ways, one corresponds to vehicle-lane interaction and the other to vehicle-vehicle interaction. The context vectors are also used to model the weights via the proposed mode selection network. To evaluate our forecasting model, we use two large-scale real-world datasets. Experimental results show that our model is not only capable of generating clear multi-modal trajectory distributions but also outperforms the state-of-the-art (SOTA) models in terms of prediction accuracy.

Chongyang Zhong, Lei Hu, Zihao Zhang, Shihong Xia

Motion capture data is largely needed in the movie and game industry in recent years. Since the motion capture system is expensive and requires manual post-processing, motion synthesis is a plausible solution to acquire more motion data. However, generating the action-conditioned, realistic, and diverse 3D human motions given the semantic action labels is still challenging because the mapping from semantic labels to real motion sequences is hard to depict. Previous work made some positive attempts like appending label tokens to pose encoding and performing action bias on latent space, however, how to synthesize diverse motions that accurately match the given label is still not fully explored. In this paper, we propose the Uncoupled-Modulation Conditional Variational AutoEncoder(UM-CVAE) to generate action-conditioned motions from scratch in an uncoupled manner. The main idea is twofold: (i)training an action-agnostic encoder to eliminate the action-related information to learn the easy-modulated latent representation; (ii)strengthening the action-conditioned process with FiLM-based action-aware modulation. We conduct extensive experiments on the HumanAct12, UESTC, and BABEL datasets, demonstrating that our method achieves state-of-the-art performance both qualitatively and quantitatively with potential applications.

Haoquan Li, Laoming Zhang, Daoan Zhang, Lang Fu, Peng Yang, Jianguo Zhang

This paper presents a transformer framework for few-shot learning, termed TransVLAD, with one focus showing the power of locally aggregated descriptors for few-shot learning. Our TransVLAD model is simple: a standard transformer encoder following a NeXtVLAD aggregation module to output the locally aggregated descriptors. In contrast to the prevailing use of CNN as part of the feature extractor, we are the first to prove self-supervised learning like masked autoencoders (MAE) can deal with the overfitting of transformers in few-shot image classification. Besides, few-shot learning can benefit from this general-purpose pre-training. Then, we propose two methods to mitigate few-shot biases, supervision bias and simple-characteristic bias. The first method is introducing masking operation into fine-tuning, by which we accelerate fine-tuning (by more than 3x) and improve accuracy. The second one is adapting focal loss into soft focal loss to focus on hard characteristics learning. Our TransVLAD finally tops 10 benchmarks on five popular few-shot datasets by an average of more than 2%.

Zhenqi Fu, Wu Wang, Yue Huang, Xinghao Ding, Kai-Kuang Ma

A main challenge faced in the deep learning-based Underwater Image Enhancement (UIE) is that the ground truth high-quality image is unavailable. Most of the existing methods first generate approximate reference maps and then train an enhancement network with certainty. This kind of method fails to handle the ambiguity of the reference map. In this paper, we resolve UIE into distribution estimation and consensus process. We present a novel probabilistic network to learn the enhancement distribution of degraded underwater images. Specifically, we combine conditional variational autoencoder with adaptive instance normalization to construct the enhancement distribution. After that, we adopt a consensus process to predict a deterministic result based on a set of samples from the distribution. By learning the enhancement distribution, our method can cope with the bias introduced in the reference map labeling to some extent. Additionally, the consensus process is useful to capture a robust and stable result. We examined the proposed method on two widely used real-world underwater image enhancement datasets. Experimental results demonstrate that our approach enables sampling possible enhancement predictions. Meanwhile, the consensus estimate yields competitive performance compared with state-of-the-art UIE methods.

Yung-Han Ho, Chih-Peng Chang, Peng-Yu Chen, Alessandro Gnutti, Wen-Hsiao Peng

This paper presents an end-to-end learning-based video compression system, termed CANF-VC, based on conditional augmented normalizing flows (CANF). Most learned video compression systems adopt the same hybrid-based coding architecture as the traditional codecs. Recent research on conditional coding has shown the sub-optimality of the hybrid-based coding and opens up opportunities for deep generative models to take a key role in creating new coding frameworks. CANF-VC represents a new attempt that leverages the conditional ANF to learn a video generative model for conditional inter-frame coding. We choose ANF because it is a special type of generative model, which includes variational autoencoder as a special case and is able to achieve better expressiveness. CANF-VC also extends the idea of conditional coding to motion coding, forming a purely conditional coding framework. Extensive experimental results on commonly used datasets confirm the superiority of CANF-VC to the state-of-the-art methods. The source code of CANF-VC is available at https://github.com/NYCU-MAPL/CANF-VC.

Zhanghan Ke, Chunyi Sun, Lei Zhu, Ke Xu, Rynson W.H. Lau

Recent works on image harmonization solve the problem as a pixel-wise image translation task via large autoencoders. They have unsatisfactory performances and slow inference speeds when dealing with high-resolution images. In this work, we observe that adjusting the input arguments of basic image filters, e.g., brightness and contrast, is sufficient for humans to produce realistic images from the composite ones. Hence, we frame image harmonization as an image-level regression problem to learn the arguments of the filters that humans use for the task. We present a Harmonizer framework for image harmonization. Unlike prior methods that are based on black-box autoencoders, Harmonizer contains a neural network for filter argument prediction and several white-box filters (based on the predicted arguments) for image harmonization. We also introduce a cascade regressor and a dynamic loss strategy for Harmonizer to learn filter arguments more stably and precisely. Since our network only outputs image-level arguments and the filters we used are efficient, Harmonizer is much lighter and faster than existing methods. Comprehensive experiments demonstrate that Harmonizer surpasses existing methods notably, especially with high-resolution inputs. Finally, we apply Harmonizer to video harmonization, which achieves consistent results across frames and 56 fps at 1080P resolution.

Chenjie Cao, Qiaole Dong, Yanwei Fu

Many recent inpainting works have achieved impressive results by leveraging Deep Neural Networks (DNNs) to model various prior information for image restoration. Unfortunately, the performance of these methods is largely limited by the representation ability of vanilla Convolutional Neural Networks (CNNs) backbones. On the other hand, Vision Transformers (ViT) with self-supervised pre-training have shown great potential for many visual recognition and object detection tasks. A natural question is whether the inpainting task can be greatly benefited from the ViT backbone? However, it is nontrivial to directly replace the new backbones in inpainting networks, as the inpainting is an inverse problem fundamentally different from the recognition tasks. To this end, this paper incorporates the pre-training based Masked AutoEncoder (MAE) into the inpainting model, which enjoys richer informative priors to enhance the inpainting process. Moreover, we propose to use attention priors from MAE to make the inpainting model learn more long-distance dependencies between masked and unmasked regions. Sufficient ablations have been discussed about the inpainting and the self-supervised pre-training models in this paper. Besides, experiments on both Places2 and FFHQ demonstrate the effectiveness of our proposed model. Codes and pre-trained models will be released.