Zhenyu Zhang

Tianlong Chen, Zhenyu Zhang, Yu Cheng et al.

Vision transformers (ViTs) have gained increasing popularity as they are commonly believed to own higher modeling capacity and representation flexibility, than traditional convolutional networks. However, it is questionable whether such potential has been fully unleashed in practice, as the learned ViTs often suffer from over-smoothening, yielding likely redundant models. Recent works made preliminary attempts to identify and alleviate such redundancy, e.g., via regularizing embedding similarity or re-injecting convolution-like structures. However, a "head-to-toe assessment" regarding the extent of redundancy in ViTs, and how much we could gain by thoroughly mitigating such, has been absent for this field. This paper, for the first time, systematically studies the ubiquitous existence of redundancy at all three levels: patch embedding, attention map, and weight space. In view of them, we advocate a principle of diversity for training ViTs, by presenting corresponding regularizers that encourage the representation diversity and coverage at each of those levels, that enabling capturing more discriminative information. Extensive experiments on ImageNet with a number of ViT backbones validate the effectiveness of our proposals, largely eliminating the observed ViT redundancy and significantly boosting the model generalization. For example, our diversified DeiT obtains 0.70%~1.76% accuracy boosts on ImageNet with highly reduced similarity. Our codes are fully available in https://github.com/VITA-Group/Diverse-ViT.

Qiming Peng, Yinxu Pan, Wenjin Wang et al.

Recent years have witnessed the rise and success of pre-training techniques in visually-rich document understanding. However, most existing methods lack the systematic mining and utilization of layout-centered knowledge, leading to sub-optimal performances. In this paper, we propose ERNIE-Layout, a novel document pre-training solution with layout knowledge enhancement in the whole workflow, to learn better representations that combine the features from text, layout, and image. Specifically, we first rearrange input sequences in the serialization stage, and then present a correlative pre-training task, reading order prediction, to learn the proper reading order of documents. To improve the layout awareness of the model, we integrate a spatial-aware disentangled attention into the multi-modal transformer and a replaced regions prediction task into the pre-training phase. Experimental results show that ERNIE-Layout achieves superior performance on various downstream tasks, setting new state-of-the-art on key information extraction, document image classification, and document question answering datasets. The code and models are publicly available at http://github.com/PaddlePaddle/PaddleNLP/tree/develop/model_zoo/ernie-layout.

Lu Yin, You Wu, Zhenyu Zhang et al.

Large Language Models (LLMs), renowned for their remarkable performance across diverse domains, present a challenge when it comes to practical deployment due to their colossal model size. In response to this challenge, efforts have been directed toward the application of traditional network pruning techniques to LLMs, uncovering a massive number of parameters that can be pruned in one-shot without hurting performance. Prevailing LLM pruning strategies have consistently adhered to the practice of uniformly pruning all layers at equivalent sparsity, resulting in robust performance. However, this observation stands in contrast to the prevailing trends observed in the field of vision models, where non-uniform layerwise sparsity typically yields stronger results. To understand the underlying reasons for this disparity, we conduct a comprehensive study and discover a strong correlation with the emergence of activation outliers in LLMs. Inspired by this finding, we introduce a novel LLM pruning methodology that incorporates a tailored set of non-uniform layerwise sparsity ratios, termed as Outlier Weighed Layerwise sparsity (OWL). The sparsity ratio of OWL is proportional to the outlier ratio observed within each layer, facilitating a more effective alignment between layerwise weight sparsity and outlier ratios. Our empirical evaluation, conducted across the LLaMA-V1 family and OPT, spanning various benchmarks, demonstrates the distinct advantages offered by OWL over previous methods. For instance, OWL exhibits a remarkable performance gain, surpassing the state-of-the-art Wanda and SparseGPT by 61.22 and 6.80 perplexity at a high sparsity level of 70%, respectively, while delivering 2.6x end-to-end inference speed-up in the DeepSparse inference engine. Codes are available at https://github.com/luuyin/OWL.

Yilong Chen, Guoxia Wang, Junyuan Shang et al.

Large Language Models (LLMs) have ignited an innovative surge of AI applications, marking a new era of exciting possibilities equipped with extended context windows. However, hosting these models is cost-prohibitive mainly due to the extensive memory consumption of KV Cache involving long-context modeling. Despite several works proposing to evict unnecessary tokens from the KV Cache, most of them rely on the biased local statistics of accumulated attention scores and report performance using unconvincing metric like perplexity on inadequate short-text evaluation. In this paper, we propose NACL, a general framework for long-context KV cache eviction that achieves more optimal and efficient eviction in a single operation during the encoding phase. Due to NACL's efficiency, we combine more accurate attention score statistics in PROXY TOKENS EVICTION with the diversified random eviction strategy of RANDOM EVICTION, aiming to alleviate the issue of attention bias and enhance the robustness in maintaining pivotal tokens for long-context modeling tasks. Notably, our method significantly improves the performance on short- and long-text tasks by 80% and 76% respectively, reducing KV Cache by up to 50% with over 95% performance maintenance. The code is available at https://github.com/PaddlePaddle/Research/tree/master/NLP/ACL2024-NACL.

Shiwei Liu, Tianlong Chen, Zhenyu Zhang et al.

Sparse Neural Networks (SNNs) have received voluminous attention predominantly due to growing computational and memory footprints of consistently exploding parameter count in large-scale models. Similar to their dense counterparts, recent SNNs generalize just as well and are equipped with numerous favorable benefits (e.g., low complexity, high scalability, and robustness), sometimes even better than the original dense networks. As research effort is focused on developing increasingly sophisticated sparse algorithms, it is startling that a comprehensive benchmark to evaluate the effectiveness of these algorithms has been highly overlooked. In absence of a carefully crafted evaluation benchmark, most if not all, sparse algorithms are evaluated against fairly simple and naive tasks (eg. CIFAR, ImageNet, GLUE, etc.), which can potentially camouflage many advantages as well unexpected predicaments of SNNs. In pursuit of a more general evaluation and unveiling the true potential of sparse algorithms, we introduce "Sparsity May Cry" Benchmark (SMC-Bench), a collection of carefully-curated 4 diverse tasks with 10 datasets, that accounts for capturing a wide range of domain-specific and sophisticated knowledge. Our systemic evaluation of the most representative sparse algorithms reveals an important obscured observation: the state-of-the-art magnitude- and/or gradient-based sparse algorithms seemingly fail to perform on SMC-Bench when applied out-of-the-box, sometimes at significantly trivial sparsity as low as 5%. By incorporating these well-thought and diverse tasks, SMC-Bench is designed to favor and encourage the development of more scalable and generalizable sparse algorithms.

Ruisi Cai, Zhenyu Zhang, Zhangyang Wang

Given a robust model trained to be resilient to one or multiple types of distribution shifts (e.g., natural image corruptions), how is that "robustness" encoded in the model weights, and how easily can it be disentangled and/or "zero-shot" transferred to some other models? This paper empirically suggests a surprisingly simple answer: linearly - by straightforward model weight arithmetic! We start by drawing several key observations: (1)assuming that we train the same model architecture on both a clean dataset and its corrupted version, resultant weights mostly differ in shallow layers; (2)the weight difference after projection, which we call "Robust Weight Signature" (RWS), appears to be discriminative and indicative of different corruption types; (3)for the same corruption type, the RWSs obtained by one model architecture are highly consistent and transferable across different datasets. We propose a minimalistic model robustness "patching" framework that carries a model trained on clean data together with its pre-extracted RWSs. In this way, injecting certain robustness to the model is reduced to directly adding the corresponding RWS to its weight. We verify our proposed framework to be remarkably (1)lightweight. since RWSs concentrate on the shallowest few layers and we further show they can be painlessly quantized, storing an RWS is up to 13 x more compact than storing the full weight copy; (2)in-situ adjustable. RWSs can be appended as needed and later taken off to restore the intact clean model. We further demonstrate one can linearly re-scale the RWS to control the patched robustness strength; (3)composable. Multiple RWSs can be added simultaneously to patch more comprehensive robustness at once; and (4)transferable. Even when the clean model backbone is continually adapted or updated, RWSs remain as effective patches due to their outstanding cross-dataset transferability.

Zhenyu Zhang, Yuming Zhao, Meng Chen et al.

Contrastive learning (CL) has achieved astonishing progress in computer vision, speech, and natural language processing fields recently with self-supervised learning. However, CL approach to the supervised setting is not fully explored, especially for the natural language understanding classification task. Intuitively, the class label itself has the intrinsic ability to perform hard positive/negative mining, which is crucial for CL. Motivated by this, we propose a novel label anchored contrastive learning approach (denoted as LaCon) for language understanding. Specifically, three contrastive objectives are devised, including a multi-head instance-centered contrastive loss (ICL), a label-centered contrastive loss (LCL), and a label embedding regularizer (LER). Our approach does not require any specialized network architecture or any extra data augmentation, thus it can be easily plugged into existing powerful pre-trained language models. Compared to the state-of-the-art baselines, LaCon obtains up to 4.1% improvement on the popular datasets of GLUE and CLUE benchmarks. Besides, LaCon also demonstrates significant advantages under the few-shot and data imbalance settings, which obtains up to 9.4% improvement on the FewGLUE and FewCLUE benchmarking tasks.

Xiaopeng Sun, Weiqi Li, Zhenyu Zhang et al.

360° omnidirectional images have gained research attention due to their immersive and interactive experience, particularly in AR/VR applications. However, they suffer from lower angular resolution due to being captured by fisheye lenses with the same sensor size for capturing planar images. To solve the above issues, we propose a two-stage framework for 360° omnidirectional image superresolution. The first stage employs two branches: model A, which incorporates omnidirectional position-aware deformable blocks (OPDB) and Fourier upsampling, and model B, which adds a spatial frequency fusion module (SFF) to model A. Model A aims to enhance the feature extraction ability of 360° image positional information, while Model B further focuses on the high-frequency information of 360° images. The second stage performs same-resolution enhancement based on the structure of model A with a pixel unshuffle operation. In addition, we collected data from YouTube to improve the fitting ability of the transformer, and created pseudo low-resolution images using a degradation network. Our proposed method achieves superior performance and wins the NTIRE 2023 challenge of 360° omnidirectional image super-resolution.

Yang Wu, Pai Peng, Zhenyu Zhang et al.

Recent works on multi-modal emotion recognition move towards end-to-end models, which can extract the task-specific features supervised by the target task compared with the two-phase pipeline. However, previous methods only model the feature interactions between the textual and either acoustic and visual modalities, ignoring capturing the feature interactions between the acoustic and visual modalities. In this paper, we propose the multi-modal end-to-end transformer (ME2ET), which can effectively model the tri-modal features interaction among the textual, acoustic, and visual modalities at the low-level and high-level. At the low-level, we propose the progressive tri-modal attention, which can model the tri-modal feature interactions by adopting a two-pass strategy and can further leverage such interactions to significantly reduce the computation and memory complexity through reducing the input token length. At the high-level, we introduce the tri-modal feature fusion layer to explicitly aggregate the semantic representations of three modalities. The experimental results on the CMU-MOSEI and IEMOCAP datasets show that ME2ET achieves the state-of-the-art performance. The further in-depth analysis demonstrates the effectiveness, efficiency, and interpretability of the proposed progressive tri-modal attention, which can help our model to achieve better performance while significantly reducing the computation and memory cost. Our code will be publicly available.

Xiantao Hu, Ying Tai, Xu Zhao et al.

Multimodal tracking has garnered widespread attention as a result of its ability to effectively address the inherent limitations of traditional RGB tracking. However, existing multimodal trackers mainly focus on the fusion and enhancement of spatial features or merely leverage the sparse temporal relationships between video frames. These approaches do not fully exploit the temporal correlations in multimodal videos, making it difficult to capture the dynamic changes and motion information of targets in complex scenarios. To alleviate this problem, we propose a unified multimodal spatial-temporal tracking approach named STTrack. In contrast to previous paradigms that solely relied on updating reference information, we introduced a temporal state generator (TSG) that continuously generates a sequence of tokens containing multimodal temporal information. These temporal information tokens are used to guide the localization of the target in the next time state, establish long-range contextual relationships between video frames, and capture the temporal trajectory of the target. Furthermore, at the spatial level, we introduced the mamba fusion and background suppression interactive (BSI) modules. These modules establish a dual-stage mechanism for coordinating information interaction and fusion between modalities. Extensive comparisons on five benchmark datasets illustrate that STTrack achieves state-of-the-art performance across various multimodal tracking scenarios. Code is available at: https://github.com/NJU-PCALab/STTrack.

Zhenyu Zhang, Guangyao Chen, Yixiong Zou et al.

Humans exhibit a remarkable ability to learn quickly from a limited number of labeled samples, a capability that starkly contrasts with that of current machine learning systems. Unsupervised Few-Shot Learning (U-FSL) seeks to bridge this divide by reducing reliance on annotated datasets during initial training phases. In this work, we first quantitatively assess the impacts of Masked Image Modeling (MIM) and Contrastive Learning (CL) on few-shot learning tasks. Our findings highlight the respective limitations of MIM and CL in terms of discriminative and generalization abilities, which contribute to their underperformance in U-FSL contexts. To address these trade-offs between generalization and discriminability in unsupervised pretraining, we introduce a novel paradigm named Masked Image Contrastive Modeling (MICM). MICM creatively combines the targeted object learning strength of CL with the generalized visual feature learning capability of MIM, significantly enhancing its efficacy in downstream few-shot learning inference. Extensive experimental analyses confirm the advantages of MICM, demonstrating significant improvements in both generalization and discrimination capabilities for few-shot learning. Our comprehensive quantitative evaluations further substantiate the superiority of MICM, showing that our two-stage U-FSL framework based on MICM markedly outperforms existing leading baselines.

Zhaocong liu, Fa Zhang, Lin Cheng et al.

High-quality labeled datasets are essential for deep learning. Traditional manual annotation methods are not only costly and inefficient but also pose challenges in specialized domains where expert knowledge is needed. Self-supervised methods, despite leveraging unlabeled data for feature extraction, still require hundreds or thousands of labeled instances to guide the model for effective specialized image classification. Current unsupervised learning methods offer automatic classification without prior annotation but often compromise on accuracy. As a result, efficiently procuring high-quality labeled datasets remains a pressing challenge for specialized domain images devoid of annotated data. Addressing this, an unsupervised classification method with three key ideas is introduced: 1) dual-step feature dimensionality reduction using a pre-trained model and manifold learning, 2) a voting mechanism from multiple clustering algorithms, and 3) post-hoc instead of prior manual annotation. This approach outperforms supervised methods in classification accuracy, as demonstrated with fungal image data, achieving 94.1% and 96.7% on public and private datasets respectively. The proposed unsupervised classification method reduces dependency on pre-annotated datasets, enabling a closed-loop for data classification. The simplicity and ease of use of this method will also bring convenience to researchers in various fields in building datasets, promoting AI applications for images in specialized domains.

Tianjin Huang, Lu Yin, Zhenyu Zhang et al.

This paper reveals a new appeal of the recently emerged large-kernel Convolutional Neural Networks (ConvNets): as the teacher in Knowledge Distillation (KD) for small-kernel ConvNets. While Transformers have led state-of-the-art (SOTA) performance in various fields with ever-larger models and labeled data, small-kernel ConvNets are considered more suitable for resource-limited applications due to the efficient convolution operation and compact weight sharing. KD is widely used to boost the performance of small-kernel ConvNets. However, previous research shows that it is not quite effective to distill knowledge (e.g., global information) from Transformers to small-kernel ConvNets, presumably due to their disparate architectures. We hereby carry out a first-of-its-kind study unveiling that modern large-kernel ConvNets, a compelling competitor to Vision Transformers, are remarkably more effective teachers for small-kernel ConvNets, due to more similar architectures. Our findings are backed up by extensive experiments on both logit-level and feature-level KD ``out of the box", with no dedicated architectural nor training recipe modifications. Notably, we obtain the \textbf{best-ever pure ConvNet} under 30M parameters with \textbf{83.1\%} top-1 accuracy on ImageNet, outperforming current SOTA methods including ConvNeXt V2 and Swin V2. We also find that beneficial characteristics of large-kernel ConvNets, e.g., larger effective receptive fields, can be seamlessly transferred to students through this large-to-small kernel distillation. Code is available at: \url{https://github.com/VITA-Group/SLaK}.

Xuxi Chen, Zhenyu Zhang, Yongduo Sui et al.

Deep generative adversarial networks (GANs) have gained growing popularity in numerous scenarios, while usually suffer from high parameter complexities for resource-constrained real-world applications. However, the compression of GANs has less been explored. A few works show that heuristically applying compression techniques normally leads to unsatisfactory results, due to the notorious training instability of GANs. In parallel, the lottery ticket hypothesis shows prevailing success on discriminative models, in locating sparse matching subnetworks capable of training in isolation to full model performance. In this work, we for the first time study the existence of such trainable matching subnetworks in deep GANs. For a range of GANs, we certainly find matching subnetworks at 67%-74% sparsity. We observe that with or without pruning discriminator has a minor effect on the existence and quality of matching subnetworks, while the initialization weights used in the discriminator play a significant role. We then show the powerful transferability of these subnetworks to unseen tasks. Furthermore, extensive experimental results demonstrate that our found subnetworks substantially outperform previous state-of-the-art GAN compression approaches in both image generation (e.g. SNGAN) and image-to-image translation GANs (e.g. CycleGAN). Codes available at https://github.com/VITA-Group/GAN-LTH.

Ren Yang, Radu Timofte, Jing Liu et al.

This paper reviews the first NTIRE challenge on quality enhancement of compressed video, with a focus on the proposed methods and results. In this challenge, the new Large-scale Diverse Video (LDV) dataset is employed. The challenge has three tracks. Tracks 1 and 2 aim at enhancing the videos compressed by HEVC at a fixed QP, while Track 3 is designed for enhancing the videos compressed by x265 at a fixed bit-rate. Besides, the quality enhancement of Tracks 1 and 3 targets at improving the fidelity (PSNR), and Track 2 targets at enhancing the perceptual quality. The three tracks totally attract 482 registrations. In the test phase, 12 teams, 8 teams and 11 teams submitted the final results of Tracks 1, 2 and 3, respectively. The proposed methods and solutions gauge the state-of-the-art of video quality enhancement. The homepage of the challenge: https://github.com/RenYang-home/NTIRE21_VEnh

Tianlong Chen, Zhenyu Zhang, Xu Ouyang et al.

Batch normalization (BN) is a key facilitator and considered essential for state-of-the-art binary neural networks (BNN). However, the BN layer is costly to calculate and is typically implemented with non-binary parameters, leaving a hurdle for the efficient implementation of BNN training. It also introduces undesirable dependence between samples within each batch. Inspired by the latest advance on Batch Normalization Free (BN-Free) training, we extend their framework to training BNNs, and for the first time demonstrate that BNs can be completed removed from BNN training and inference regimes. By plugging in and customizing techniques including adaptive gradient clipping, scale weight standardization, and specialized bottleneck block, a BN-free BNN is capable of maintaining competitive accuracy compared to its BN-based counterpart. Extensive experiments validate the effectiveness of our proposal across diverse BNN backbones and datasets. For example, after removing BNs from the state-of-the-art ReActNets, it can still be trained with our proposed methodology to achieve 92.08%, 68.34%, and 68.0% accuracy on CIFAR-10, CIFAR-100, and ImageNet respectively, with marginal performance drop (0.23%~0.44% on CIFAR and 1.40% on ImageNet). Codes and pre-trained models are available at: https://github.com/VITA-Group/BNN_NoBN.

Gan Pei, Jiangning Zhang, Menghan Hu et al.

Deepfake is a technology dedicated to creating highly realistic facial images and videos under specific conditions, which has significant application potential in fields such as entertainment, movie production, digital human creation, to name a few. With the advancements in deep learning, techniques primarily represented by Variational Autoencoders and Generative Adversarial Networks have achieved impressive generation results. More recently, the emergence of diffusion models with powerful generation capabilities has sparked a renewed wave of research. In addition to deepfake generation, corresponding detection technologies continuously evolve to regulate the potential misuse of deepfakes, such as for privacy invasion and phishing attacks. This survey comprehensively reviews the latest developments in deepfake generation and detection, summarizing and analyzing current state-of-the-arts in this rapidly evolving field. We first unify task definitions, comprehensively introduce datasets and metrics, and discuss developing technologies. Then, we discuss the development of several related sub-fields and focus on researching four representative deepfake fields: face swapping, face reenactment, talking face generation, and facial attribute editing, as well as forgery detection. Subsequently, we comprehensively benchmark representative methods on popular datasets for each field, fully evaluating the latest and influential published works. Finally, we analyze challenges and future research directions of the discussed fields.

Kui Huang, Mengke Song, Shuo Ba et al.

Waste classification is crucial for improving processing efficiency and reducing environmental pollution. Supervised deep learning methods are commonly used for automated waste classification, but they rely heavily on large labeled datasets, which are costly and inefficient to obtain. Real-world waste data often exhibit category and style biases, such as variations in camera angles, lighting conditions, and types of waste, which can impact the model's performance and generalization ability. Therefore, constructing a bias-free dataset is essential. Manual labeling is not only costly but also inefficient. While self-supervised learning helps address data scarcity, it still depends on some labeled data and generally results in lower accuracy compared to supervised methods. Unsupervised methods show potential in certain cases but typically do not perform as well as supervised models, highlighting the need for an efficient and cost-effective unsupervised approach. This study presents a novel unsupervised method, Dual-Encoder Contrastive Learning with Multi-Clustering Voting (DECMCV). The approach involves using a pre-trained ConvNeXt model for image encoding, leveraging VisionTransformer to generate positive samples, and applying a multi-clustering voting mechanism to address data labeling and domain shift issues. Experimental results demonstrate that DECMCV achieves classification accuracies of 93.78% and 98.29% on the TrashNet and Huawei Cloud datasets, respectively, outperforming or matching supervised models. On a real-world dataset of 4,169 waste images, only 50 labeled samples were needed to accurately label thousands, improving classification accuracy by 29.85% compared to supervised models. This method effectively addresses style differences, enhances model generalization, and contributes to the advancement of automated waste classification.

Lei Liu, Yuchao Lu, Ling An et al.

As human activities intensify, environmental systems such as aquatic ecosystems and water treatment systems face increasingly complex pressures, impacting ecological balance, public health, and sustainable development, making intelligent anomaly monitoring essential. However, traditional monitoring methods suffer from delayed responses, insufficient data processing capabilities, and weak generalisation, making them unsuitable for complex environmental monitoring needs.In recent years, machine learning has been widely applied to anomaly detection, but the multi-dimensional features and spatiotemporal dynamics of environmental ecological data, especially the long-term dependencies and strong variability in the time dimension, limit the effectiveness of traditional methods.Deep learning, with its ability to automatically learn features, captures complex nonlinear relationships, improving detection performance. However, its application in environmental monitoring is still in its early stages and requires further exploration.This paper introduces a new deep learning method, Time-EAPCR (Time-Embedding-Attention-Permutated CNN-Residual), and applies it to environmental science. The method uncovers feature correlations, captures temporal evolution patterns, and enables precise anomaly detection in environmental systems.We validated Time-EAPCR's high accuracy and robustness across four publicly available environmental datasets. Experimental results show that the method efficiently handles multi-source data, improves detection accuracy, and excels across various scenarios with strong adaptability and generalisation. Additionally, a real-world river monitoring dataset confirmed the feasibility of its deployment, providing reliable technical support for environmental monitoring.

Ming Cheng, Haoyu Ma, Qiufang Ma et al.

Multi-stage strategies are frequently employed in image restoration tasks. While transformer-based methods have exhibited high efficiency in single-image super-resolution tasks, they have not yet shown significant advantages over CNN-based methods in stereo super-resolution tasks. This can be attributed to two key factors: first, current single-image super-resolution transformers are unable to leverage the complementary stereo information during the process; second, the performance of transformers is typically reliant on sufficient data, which is absent in common stereo-image super-resolution algorithms. To address these issues, we propose a Hybrid Transformer and CNN Attention Network (HTCAN), which utilizes a transformer-based network for single-image enhancement and a CNN-based network for stereo information fusion. Furthermore, we employ a multi-patch training strategy and larger window sizes to activate more input pixels for super-resolution. We also revisit other advanced techniques, such as data augmentation, data ensemble, and model ensemble to reduce overfitting and data bias. Finally, our approach achieved a score of 23.90dB and emerged as the winner in Track 1 of the NTIRE 2023 Stereo Image Super-Resolution Challenge.