Yonghong Tian

Yifan Zhao, Jia Li, Xiaowu Chen et al. · pku

Fine-grained visual recognition is to classify objects with visually similar appearances into subcategories, which has made great progress with the development of deep CNNs. However, handling subtle differences between different subcategories still remains a challenge. In this paper, we propose to solve this issue in one unified framework from two aspects, i.e., constructing feature-level interrelationships, and capturing part-level discriminative features. This framework, namely PArt-guided Relational Transformers (PART), is proposed to learn the discriminative part features with an automatic part discovery module, and to explore the intrinsic correlations with a feature transformation module by adapting the Transformer models from the field of natural language processing. The part discovery module efficiently discovers the discriminative regions which are highly-corresponded to the gradient descent procedure. Then the second feature transformation module builds correlations within the global embedding and multiple part embedding, enhancing spatial interactions among semantic pixels. Moreover, our proposed approach does not rely on additional part branches in the inference time and reaches state-of-the-art performance on 3 widely-used fine-grained object recognition benchmarks. Experimental results and explainable visualizations demonstrate the effectiveness of our proposed approach. The code can be found at https://github.com/iCVTEAM/PART.

Dianze Li, Jianing Li, Yonghong Tian · pku

DAVIS camera, streaming two complementary sensing modalities of asynchronous events and frames, has gradually been used to address major object detection challenges (e.g., fast motion blur and low-light). However, how to effectively leverage rich temporal cues and fuse two heterogeneous visual streams remains a challenging endeavor. To address this challenge, we propose a novel streaming object detector with Transformer, namely SODFormer, which first integrates events and frames to continuously detect objects in an asynchronous manner. Technically, we first build a large-scale multimodal neuromorphic object detection dataset (i.e., PKU-DAVIS-SOD) over 1080.1k manual labels. Then, we design a spatiotemporal Transformer architecture to detect objects via an end-to-end sequence prediction problem, where the novel temporal Transformer module leverages rich temporal cues from two visual streams to improve the detection performance. Finally, an asynchronous attention-based fusion module is proposed to integrate two heterogeneous sensing modalities and take complementary advantages from each end, which can be queried at any time to locate objects and break through the limited output frequency from synchronized frame-based fusion strategies. The results show that the proposed SODFormer outperforms four state-of-the-art methods and our eight baselines by a significant margin. We also show that our unifying framework works well even in cases where the conventional frame-based camera fails, e.g., high-speed motion and low-light conditions. Our dataset and code can be available at https://github.com/dianzl/SODFormer.

Xiao Wang, Guangyao Chen, Guangwu Qian et al.

With the urgent demand for generalized deep models, many pre-trained big models are proposed, such as BERT, ViT, GPT, etc. Inspired by the success of these models in single domains (like computer vision and natural language processing), the multi-modal pre-trained big models have also drawn more and more attention in recent years. In this work, we give a comprehensive survey of these models and hope this paper could provide new insights and helps fresh researchers to track the most cutting-edge works. Specifically, we firstly introduce the background of multi-modal pre-training by reviewing the conventional deep learning, pre-training works in natural language process, computer vision, and speech. Then, we introduce the task definition, key challenges, and advantages of multi-modal pre-training models (MM-PTMs), and discuss the MM-PTMs with a focus on data, objectives, network architectures, and knowledge enhanced pre-training. After that, we introduce the downstream tasks used for the validation of large-scale MM-PTMs, including generative, classification, and regression tasks. We also give visualization and analysis of the model parameters and results on representative downstream tasks. Finally, we point out possible research directions for this topic that may benefit future works. In addition, we maintain a continuously updated paper list for large-scale pre-trained multi-modal big models: https://github.com/wangxiao5791509/MultiModal_BigModels_Survey. This paper has been published by the journal Machine Intelligence Research (MIR), https://link.springer.com/article/10.1007/s11633-022-1410-8, DOI: 10.1007/s11633-022-1410-8, vol. 20, no. 4, pp. 447-482, 2023.

Zeyin Song, Yifan Zhao, Yujun Shi et al. · pku

Few-shot class-incremental learning (FSCIL) aims at learning to classify new classes continually from limited samples without forgetting the old classes. The mainstream framework tackling FSCIL is first to adopt the cross-entropy (CE) loss for training at the base session, then freeze the feature extractor to adapt to new classes. However, in this work, we find that the CE loss is not ideal for the base session training as it suffers poor class separation in terms of representations, which further degrades generalization to novel classes. One tempting method to mitigate this problem is to apply an additional naive supervised contrastive learning (SCL) in the base session. Unfortunately, we find that although SCL can create a slightly better representation separation among different base classes, it still struggles to separate base classes and new classes. Inspired by the observations made, we propose Semantic-Aware Virtual Contrastive model (SAVC), a novel method that facilitates separation between new classes and base classes by introducing virtual classes to SCL. These virtual classes, which are generated via pre-defined transformations, not only act as placeholders for unseen classes in the representation space, but also provide diverse semantic information. By learning to recognize and contrast in the fantasy space fostered by virtual classes, our SAVC significantly boosts base class separation and novel class generalization, achieving new state-of-the-art performance on the three widely-used FSCIL benchmark datasets. Code is available at: https://github.com/zysong0113/SAVC.

Li Ma, Peixi Peng, Guangyao Chen et al. · pku

The sensitivity of deep neural networks to compressed images hinders their usage in many real applications, which means classification networks may fail just after taking a screenshot and saving it as a compressed file. In this paper, we argue that neglected disposable coding parameters stored in compressed files could be picked up to reduce the sensitivity of deep neural networks to compressed images. Specifically, we resort to using one of the representative parameters, quantization steps, to facilitate image classification. Firstly, based on quantization steps, we propose a novel quantization aware confidence (QAC), which is utilized as sample weights to reduce the influence of quantization on network training. Secondly, we utilize quantization steps to alleviate the variance of feature distributions, where a quantization aware batch normalization (QABN) is proposed to replace batch normalization of classification networks. Extensive experiments show that the proposed method significantly improves the performance of classification networks on CIFAR-10, CIFAR-100, and ImageNet. The code is released on https://github.com/LiMaPKU/QSAM.git

Xiao Wang, Zongzhen Wu, Bo Jiang et al.

The main streams of human activity recognition (HAR) algorithms are developed based on RGB cameras which are suffered from illumination, fast motion, privacy-preserving, and large energy consumption. Meanwhile, the biologically inspired event cameras attracted great interest due to their unique features, such as high dynamic range, dense temporal but sparse spatial resolution, low latency, low power, etc. As it is a newly arising sensor, even there is no realistic large-scale dataset for HAR. Considering its great practical value, in this paper, we propose a large-scale benchmark dataset to bridge this gap, termed HARDVS, which contains 300 categories and more than 100K event sequences. We evaluate and report the performance of multiple popular HAR algorithms, which provide extensive baselines for future works to compare. More importantly, we propose a novel spatial-temporal feature learning and fusion framework, termed ESTF, for event stream based human activity recognition. It first projects the event streams into spatial and temporal embeddings using StemNet, then, encodes and fuses the dual-view representations using Transformer networks. Finally, the dual features are concatenated and fed into a classification head for activity prediction. Extensive experiments on multiple datasets fully validated the effectiveness of our model. Both the dataset and source code will be released on \url{https://github.com/Event-AHU/HARDVS}.

Chuanming Tang, Xiao Wang, Ju Huang et al.

Combining the Color and Event cameras (also called Dynamic Vision Sensors, DVS) for robust object tracking is a newly emerging research topic in recent years. Existing color-event tracking framework usually contains multiple scattered modules which may lead to low efficiency and high computational complexity, including feature extraction, fusion, matching, interactive learning, etc. In this paper, we propose a single-stage backbone network for Color-Event Unified Tracking (CEUTrack), which achieves the above functions simultaneously. Given the event points and RGB frames, we first transform the points into voxels and crop the template and search regions for both modalities, respectively. Then, these regions are projected into tokens and parallelly fed into the unified Transformer backbone network. The output features will be fed into a tracking head for target object localization. Our proposed CEUTrack is simple, effective, and efficient, which achieves over 75 FPS and new SOTA performance. To better validate the effectiveness of our model and address the data deficiency of this task, we also propose a generic and large-scale benchmark dataset for color-event tracking, termed COESOT, which contains 90 categories and 1354 video sequences. Additionally, a new evaluation metric named BOC is proposed in our evaluation toolkit to evaluate the prominence with respect to the baseline methods. We hope the newly proposed method, dataset, and evaluation metric provide a better platform for color-event-based tracking. The dataset, toolkit, and source code will be released on: \url{https://github.com/Event-AHU/COESOT}.

Xiao Wang, Yao Rong, Zongzhen Wu et al.

Event camera-based pattern recognition is a newly arising research topic in recent years. Current researchers usually transform the event streams into images, graphs, or voxels, and adopt deep neural networks for event-based classification. Although good performance can be achieved on simple event recognition datasets, however, their results may be still limited due to the following two issues. Firstly, they adopt spatial sparse event streams for recognition only, which may fail to capture the color and detailed texture information well. Secondly, they adopt either Spiking Neural Networks (SNN) for energy-efficient recognition with suboptimal results, or Artificial Neural Networks (ANN) for energy-intensive, high-performance recognition. However, seldom of them consider achieving a balance between these two aspects. In this paper, we formally propose to recognize patterns by fusing RGB frames and event streams simultaneously and propose a new RGB frame-event recognition framework to address the aforementioned issues. The proposed method contains four main modules, i.e., memory support Transformer network for RGB frame encoding, spiking neural network for raw event stream encoding, multi-modal bottleneck fusion module for RGB-Event feature aggregation, and prediction head. Due to the scarce of RGB-Event based classification dataset, we also propose a large-scale PokerEvent dataset which contains 114 classes, and 27102 frame-event pairs recorded using a DVS346 event camera. Extensive experiments on two RGB-Event based classification datasets fully validated the effectiveness of our proposed framework. We hope this work will boost the development of pattern recognition by fusing RGB frames and event streams. Both our dataset and source code of this work will be released at https://github.com/Event-AHU/SSTFormer

Xu Liu, Jianing Li, Xiaopeng Fan et al.

Event cameras, offering high temporal resolutions and high dynamic ranges, have brought a new perspective to address common challenges (e.g., motion blur and low light) in monocular depth estimation. However, how to effectively exploit the sparse spatial information and rich temporal cues from asynchronous events remains a challenging endeavor. To this end, we propose a novel event-based monocular depth estimator with recurrent transformers, namely EReFormer, which is the first pure transformer with a recursive mechanism to process continuous event streams. Technically, for spatial modeling, a novel transformer-based encoder-decoder with a spatial transformer fusion module is presented, having better global context information modeling capabilities than CNN-based methods. For temporal modeling, we design a gate recurrent vision transformer unit that introduces a recursive mechanism into transformers, improving temporal modeling capabilities while alleviating the expensive GPU memory cost. The experimental results show that our EReFormer outperforms state-of-the-art methods by a margin on both synthetic and real-world datasets. We hope that our work will attract further research to develop stunning transformers in the event-based vision community. Our open-source code can be found in the supplemental material.

Yatian Pang, Wenxiao Wang, Francis E. H. Tay et al.

As a promising scheme of self-supervised learning, masked autoencoding has significantly advanced natural language processing and computer vision. Inspired by this, we propose a neat scheme of masked autoencoders for point cloud self-supervised learning, addressing the challenges posed by point cloud's properties, including leakage of location information and uneven information density. Concretely, we divide the input point cloud into irregular point patches and randomly mask them at a high ratio. Then, a standard Transformer based autoencoder, with an asymmetric design and a shifting mask tokens operation, learns high-level latent features from unmasked point patches, aiming to reconstruct the masked point patches. Extensive experiments show that our approach is efficient during pre-training and generalizes well on various downstream tasks. Specifically, our pre-trained models achieve 85.18% accuracy on ScanObjectNN and 94.04% accuracy on ModelNet40, outperforming all the other self-supervised learning methods. We show with our scheme, a simple architecture entirely based on standard Transformers can surpass dedicated Transformer models from supervised learning. Our approach also advances state-of-the-art accuracies by 1.5%-2.3% in the few-shot object classification. Furthermore, our work inspires the feasibility of applying unified architectures from languages and images to the point cloud.

Wei Fang, Zhaofei Yu, Zhaokun Zhou et al.

Vanilla spiking neurons in Spiking Neural Networks (SNNs) use charge-fire-reset neuronal dynamics, which can only be simulated serially and can hardly learn long-time dependencies. We find that when removing reset, the neuronal dynamics can be reformulated in a non-iterative form and parallelized. By rewriting neuronal dynamics without reset to a general formulation, we propose the Parallel Spiking Neuron (PSN), which generates hidden states that are independent of their predecessors, resulting in parallelizable neuronal dynamics and extremely high simulation speed. The weights of inputs in the PSN are fully connected, which maximizes the utilization of temporal information. To avoid the use of future inputs for step-by-step inference, the weights of the PSN can be masked, resulting in the masked PSN. By sharing weights across time-steps based on the masked PSN, the sliding PSN is proposed to handle sequences of varying lengths. We evaluate the PSN family on simulation speed and temporal/static data classification, and the results show the overwhelming advantage of the PSN family in efficiency and accuracy. To the best of our knowledge, this is the first study about parallelizing spiking neurons and can be a cornerstone for the spiking deep learning research. Our codes are available at \url{https://github.com/fangwei123456/Parallel-Spiking-Neuron}.

Man Yao, Jiakui Hu, Zhaokun Zhou et al.

Spiking Neural Networks (SNNs) provide an energy-efficient deep learning option due to their unique spike-based event-driven (i.e., spike-driven) paradigm. In this paper, we incorporate the spike-driven paradigm into Transformer by the proposed Spike-driven Transformer with four unique properties: 1) Event-driven, no calculation is triggered when the input of Transformer is zero; 2) Binary spike communication, all matrix multiplications associated with the spike matrix can be transformed into sparse additions; 3) Self-attention with linear complexity at both token and channel dimensions; 4) The operations between spike-form Query, Key, and Value are mask and addition. Together, there are only sparse addition operations in the Spike-driven Transformer. To this end, we design a novel Spike-Driven Self-Attention (SDSA), which exploits only mask and addition operations without any multiplication, and thus having up to $87.2\times$ lower computation energy than vanilla self-attention. Especially in SDSA, the matrix multiplication between Query, Key, and Value is designed as the mask operation. In addition, we rearrange all residual connections in the vanilla Transformer before the activation functions to ensure that all neurons transmit binary spike signals. It is shown that the Spike-driven Transformer can achieve 77.1\% top-1 accuracy on ImageNet-1K, which is the state-of-the-art result in the SNN field. The source code is available at https://github.com/BICLab/Spike-Driven-Transformer.

Feng Lin, Wenze Hu, Yaowei Wang et al.

Over the past few years, there has been growing interest in developing a broad, universal, and general-purpose computer vision system. Such systems have the potential to address a wide range of vision tasks simultaneously, without being limited to specific problems or data domains. This universality is crucial for practical, real-world computer vision applications. In this study, our focus is on a specific challenge: the large-scale, multi-domain universal object detection problem, which contributes to the broader goal of achieving a universal vision system. This problem presents several intricate challenges, including cross-dataset category label duplication, label conflicts, and the necessity to handle hierarchical taxonomies. To address these challenges, we introduce our approach to label handling, hierarchy-aware loss design, and resource-efficient model training utilizing a pre-trained large vision model. Our method has demonstrated remarkable performance, securing a prestigious second-place ranking in the object detection track of the Robust Vision Challenge 2022 (RVC 2022) on a million-scale cross-dataset object detection benchmark. We believe that our comprehensive study will serve as a valuable reference and offer an alternative approach for addressing similar challenges within the computer vision community. The source code for our work is openly available at https://github.com/linfeng93/Large-UniDet.

Qinqin Zhou, Kekai Sheng, Xiawu Zheng et al. · tencent-ai

Recently, Vision Transformer (ViT) has achieved remarkable success in several computer vision tasks. The progresses are highly relevant to the architecture design, then it is worthwhile to propose Transformer Architecture Search (TAS) to search for better ViTs automatically. However, current TAS methods are time-consuming and existing zero-cost proxies in CNN do not generalize well to the ViT search space according to our experimental observations. In this paper, for the first time, we investigate how to conduct TAS in a training-free manner and devise an effective training-free TAS (TF-TAS) scheme. Firstly, we observe that the properties of multi-head self-attention (MSA) and multi-layer perceptron (MLP) in ViTs are quite different and that the synaptic diversity of MSA affects the performance notably. Secondly, based on the observation, we devise a modular strategy in TF-TAS that evaluates and ranks ViT architectures from two theoretical perspectives: synaptic diversity and synaptic saliency, termed as DSS-indicator. With DSS-indicator, evaluation results are strongly correlated with the test accuracies of ViT models. Experimental results demonstrate that our TF-TAS achieves a competitive performance against the state-of-the-art manually or automatically design ViT architectures, and it promotes the searching efficiency in ViT search space greatly: from about $24$ GPU days to less than $0.5$ GPU days. Moreover, the proposed DSS-indicator outperforms the existing cutting-edge zero-cost approaches (e.g., TE-score and NASWOT).

Xiao Wang, Shiao Wang, Chuanming Tang et al.

Tracking using bio-inspired event cameras has drawn more and more attention in recent years. Existing works either utilize aligned RGB and event data for accurate tracking or directly learn an event-based tracker. The first category needs more cost for inference and the second one may be easily influenced by noisy events or sparse spatial resolution. In this paper, we propose a novel hierarchical knowledge distillation framework that can fully utilize multi-modal / multi-view information during training to facilitate knowledge transfer, enabling us to achieve high-speed and low-latency visual tracking during testing by using only event signals. Specifically, a teacher Transformer-based multi-modal tracking framework is first trained by feeding the RGB frame and event stream simultaneously. Then, we design a new hierarchical knowledge distillation strategy which includes pairwise similarity, feature representation, and response maps-based knowledge distillation to guide the learning of the student Transformer network. Moreover, since existing event-based tracking datasets are all low-resolution ($346 \times 260$), we propose the first large-scale high-resolution ($1280 \times 720$) dataset named EventVOT. It contains 1141 videos and covers a wide range of categories such as pedestrians, vehicles, UAVs, ping pongs, etc. Extensive experiments on both low-resolution (FE240hz, VisEvent, COESOT), and our newly proposed high-resolution EventVOT dataset fully validated the effectiveness of our proposed method. The dataset, evaluation toolkit, and source code are available on \url{https://github.com/Event-AHU/EventVOT_Benchmark}

Yifan Zhao, Tong Zhang, Jia Li et al. · pku

Few-shot learning aims to recognize novel queries with limited support samples by learning from base knowledge. Recent progress in this setting assumes that the base knowledge and novel query samples are distributed in the same domains, which are usually infeasible for realistic applications. Toward this issue, we propose to address the cross-domain few-shot learning problem where only extremely few samples are available in target domains. Under this realistic setting, we focus on the fast adaptation capability of meta-learners by proposing an effective dual adaptive representation alignment approach. In our approach, a prototypical feature alignment is first proposed to recalibrate support instances as prototypes and reproject these prototypes with a differentiable closed-form solution. Therefore feature spaces of learned knowledge can be adaptively transformed to query spaces by the cross-instance and cross-prototype relations. Besides the feature alignment, we further present a normalized distribution alignment module, which exploits prior statistics of query samples for solving the covariant shifts among the support and query samples. With these two modules, a progressive meta-learning framework is constructed to perform the fast adaptation with extremely few-shot samples while maintaining its generalization capabilities. Experimental evidence demonstrates our approach achieves new state-of-the-art results on 4 CDFSL benchmarks and 4 fine-grained cross-domain benchmarks.

Yanqi Chen, Zhengyu Ma, Wei Fang et al.

Soft threshold pruning is among the cutting-edge pruning methods with state-of-the-art performance. However, previous methods either perform aimless searching on the threshold scheduler or simply set the threshold trainable, lacking theoretical explanation from a unified perspective. In this work, we reformulate soft threshold pruning as an implicit optimization problem solved using the Iterative Shrinkage-Thresholding Algorithm (ISTA), a classic method from the fields of sparse recovery and compressed sensing. Under this theoretical framework, all threshold tuning strategies proposed in previous studies of soft threshold pruning are concluded as different styles of tuning $L_1$-regularization term. We further derive an optimal threshold scheduler through an in-depth study of threshold scheduling based on our framework. This scheduler keeps $L_1$-regularization coefficient stable, implying a time-invariant objective function from the perspective of optimization. In principle, the derived pruning algorithm could sparsify any mathematical model trained via SGD. We conduct extensive experiments and verify its state-of-the-art performance on both Artificial Neural Networks (ResNet-50 and MobileNet-V1) and Spiking Neural Networks (SEW ResNet-18) on ImageNet datasets. On the basis of this framework, we derive a family of pruning methods, including sparsify-during-training, early pruning, and pruning at initialization. The code is available at https://github.com/Yanqi-Chen/LATS.

Jianyang Zhai, Xiawu Zheng, Chang-Dong Wang et al.

Pre-trained language models (PLMs) have demonstrated strong performance in sequential recommendation (SR), which are utilized to extract general knowledge. However, existing methods still lack domain knowledge and struggle to capture users' fine-grained preferences. Meanwhile, many traditional SR methods improve this issue by integrating side information while suffering from information loss. To summarize, we believe that a good recommendation system should utilize both general and domain knowledge simultaneously. Therefore, we introduce an external knowledge base and propose Knowledge Prompt-tuning for Sequential Recommendation (\textbf{KP4SR}). Specifically, we construct a set of relationship templates and transform a structured knowledge graph (KG) into knowledge prompts to solve the problem of the semantic gap. However, knowledge prompts disrupt the original data structure and introduce a significant amount of noise. We further construct a knowledge tree and propose a knowledge tree mask, which restores the data structure in a mask matrix form, thus mitigating the noise problem. We evaluate KP4SR on three real-world datasets, and experimental results show that our approach outperforms state-of-the-art methods on multiple evaluation metrics. Specifically, compared with PLM-based methods, our method improves NDCG@5 and HR@5 by \textcolor{red}{40.65\%} and \textcolor{red}{36.42\%} on the books dataset, \textcolor{red}{11.17\%} and \textcolor{red}{11.47\%} on the music dataset, and \textcolor{red}{22.17\%} and \textcolor{red}{19.14\%} on the movies dataset, respectively. Our code is publicly available at the link: \href{https://github.com/zhaijianyang/KP4SR}{\textcolor{blue}{https://github.com/zhaijianyang/KP4SR}.}

Zhaokun Zhou, Yuesheng Zhu, Chao He et al.

We consider two biologically plausible structures, the Spiking Neural Network (SNN) and the self-attention mechanism. The former offers an energy-efficient and event-driven paradigm for deep learning, while the latter has the ability to capture feature dependencies, enabling Transformer to achieve good performance. It is intuitively promising to explore the marriage between them. In this paper, we consider leveraging both self-attention capability and biological properties of SNNs, and propose a novel Spiking Self Attention (SSA) as well as a powerful framework, named Spiking Transformer (Spikformer). The SSA mechanism in Spikformer models the sparse visual feature by using spike-form Query, Key, and Value without softmax. Since its computation is sparse and avoids multiplication, SSA is efficient and has low computational energy consumption. It is shown that Spikformer with SSA can outperform the state-of-the-art SNNs-like frameworks in image classification on both neuromorphic and static datasets. Spikformer (66.3M parameters) with comparable size to SEW-ResNet-152 (60.2M,69.26%) can achieve 74.81% top1 accuracy on ImageNet using 4 time steps, which is the state-of-the-art in directly trained SNNs models.

Yifan Zhao, Jia Li, Yonghong Tian · pku

Fine-grained visual parsing, including fine-grained part segmentation and fine-grained object recognition, has attracted considerable critical attention due to its importance in many real-world applications, e.g., agriculture, remote sensing, and space technologies. Predominant research efforts tackle these fine-grained sub-tasks following different paradigms, while the inherent relations between these tasks are neglected. Moreover, given most of the research remains fragmented, we conduct an in-depth study of the advanced work from a new perspective of learning the part relationship. In this perspective, we first consolidate recent research and benchmark syntheses with new taxonomies. Based on this consolidation, we revisit the universal challenges in fine-grained part segmentation and recognition tasks and propose new solutions by part relationship learning for these important challenges. Furthermore, we conclude several promising lines of research in fine-grained visual parsing for future research.

Wei Fang, Yanqi Chen, Jianhao Ding et al.

Spiking neural networks (SNNs) aim to realize brain-inspired intelligence on neuromorphic chips with high energy efficiency by introducing neural dynamics and spike properties. As the emerging spiking deep learning paradigm attracts increasing interest, traditional programming frameworks cannot meet the demands of the automatic differentiation, parallel computation acceleration, and high integration of processing neuromorphic datasets and deployment. In this work, we present the SpikingJelly framework to address the aforementioned dilemma. We contribute a full-stack toolkit for pre-processing neuromorphic datasets, building deep SNNs, optimizing their parameters, and deploying SNNs on neuromorphic chips. Compared to existing methods, the training of deep SNNs can be accelerated $11\times$, and the superior extensibility and flexibility of SpikingJelly enable users to accelerate custom models at low costs through multilevel inheritance and semiautomatic code generation. SpikingJelly paves the way for synthesizing truly energy-efficient SNN-based machine intelligence systems, which will enrich the ecology of neuromorphic computing.

Man Yao, Guangshe Zhao, Hengyu Zhang et al.

Benefiting from the event-driven and sparse spiking characteristics of the brain, spiking neural networks (SNNs) are becoming an energy-efficient alternative to artificial neural networks (ANNs). However, the performance gap between SNNs and ANNs has been a great hindrance to deploying SNNs ubiquitously for a long time. To leverage the full potential of SNNs, we study the effect of attention mechanisms in SNNs. We first present our idea of attention with a plug-and-play kit, termed the Multi-dimensional Attention (MA). Then, a new attention SNN architecture with end-to-end training called "MA-SNN" is proposed, which infers attention weights along the temporal, channel, as well as spatial dimensions separately or simultaneously. Based on the existing neuroscience theories, we exploit the attention weights to optimize membrane potentials, which in turn regulate the spiking response in a data-dependent way. At the cost of negligible additional parameters, MA facilitates vanilla SNNs to achieve sparser spiking activity, better performance, and energy efficiency concurrently. Experiments are conducted in event-based DVS128 Gesture/Gait action recognition and ImageNet-1k image classification. On Gesture/Gait, the spike counts are reduced by 84.9%/81.6%, and the task accuracy and energy efficiency are improved by 5.9%/4.7% and 3.4$\times$/3.2$\times$. On ImageNet-1K, we achieve top-1 accuracy of 75.92% and 77.08% on single/4-step Res-SNN-104, which are state-of-the-art results in SNNs. To our best knowledge, this is for the first time, that the SNN community achieves comparable or even better performance compared with its ANN counterpart in the large-scale dataset. Our work lights up SNN's potential as a general backbone to support various applications for SNNs, with a great balance between effectiveness and efficiency.

Haojia Lin, Xiawu Zheng, Lijiang Li et al.

Recent advances in 3D point cloud analysis bring a diverse set of network architectures to the field. However, the lack of a unified framework to interpret those networks makes any systematic comparison, contrast, or analysis challenging, and practically limits healthy development of the field. In this paper, we take the initiative to explore and propose a unified framework called PointMeta, to which the popular 3D point cloud analysis approaches could fit. This brings three benefits. First, it allows us to compare different approaches in a fair manner, and use quick experiments to verify any empirical observations or assumptions summarized from the comparison. Second, the big picture brought by PointMeta enables us to think across different components, and revisit common beliefs and key design decisions made by the popular approaches. Third, based on the learnings from the previous two analyses, by doing simple tweaks on the existing approaches, we are able to derive a basic building block, termed PointMetaBase. It shows very strong performance in efficiency and effectiveness through extensive experiments on challenging benchmarks, and thus verifies the necessity and benefits of high-level interpretation, contrast, and comparison like PointMeta. In particular, PointMetaBase surpasses the previous state-of-the-art method by 0.7%/1.4/%2.1% mIoU with only 2%/11%/13% of the computation cost on the S3DIS datasets.

Xiao Wang, Shiao Wang, Pengpeng Shao et al.

Human Action Recognition (HAR) stands as a pivotal research domain in both computer vision and artificial intelligence, with RGB cameras dominating as the preferred tool for investigation and innovation in this field. However, in real-world applications, RGB cameras encounter numerous challenges, including light conditions, fast motion, and privacy concerns. Consequently, bio-inspired event cameras have garnered increasing attention due to their advantages of low energy consumption, high dynamic range, etc. Nevertheless, most existing event-based HAR datasets are low resolution ($346 \times 260$). In this paper, we propose a large-scale, high-definition ($1280 \times 800$) human action recognition dataset based on the CeleX-V event camera, termed CeleX-HAR. It encompasses 150 commonly occurring action categories, comprising a total of 124,625 video sequences. Various factors such as multi-view, illumination, action speed, and occlusion are considered when recording these data. To build a more comprehensive benchmark dataset, we report over 20 mainstream HAR models for future works to compare. In addition, we also propose a novel Mamba vision backbone network for event stream based HAR, termed EVMamba, which equips the spatial plane multi-directional scanning and novel voxel temporal scanning mechanism. By encoding and mining the spatio-temporal information of event streams, our EVMamba has achieved favorable results across multiple datasets. Both the dataset and source code will be released on \url{https://github.com/Event-AHU/CeleX-HAR}

Yunpeng Zhai, Peixi Peng, Mengxi Jia et al. · pku

Unsupervised person re-identification has achieved great success through the self-improvement of individual neural networks. However, limited by the lack of diversity of discriminant information, a single network has difficulty learning sufficient discrimination ability by itself under unsupervised conditions. To address this limit, we develop a population-based evolutionary gaming (PEG) framework in which a population of diverse neural networks is trained concurrently through selection, reproduction, mutation, and population mutual learning iteratively. Specifically, the selection of networks to preserve is modeled as a cooperative game and solved by the best-response dynamics, then the reproduction and mutation are implemented by cloning and fluctuating hyper-parameters of networks to learn more diversity, and population mutual learning improves the discrimination of networks by knowledge distillation from each other within the population. In addition, we propose a cross-reference scatter (CRS) to approximately evaluate re-ID models without labeled samples and adopt it as the criterion of network selection in PEG. CRS measures a model's performance by indirectly estimating the accuracy of its predicted pseudo-labels according to the cohesion and separation of the feature space. Extensive experiments demonstrate that (1) CRS approximately measures the performance of models without labeled samples; (2) and PEG produces new state-of-the-art accuracy for person re-identification, indicating the great potential of population-based network cooperative training for unsupervised learning.

Yuntong Ye, Changfeng Yu, Yi Chang et al.

Image deraining is a typical low-level image restoration task, which aims at decomposing the rainy image into two distinguishable layers: the clean image layer and the rain layer. Most of the existing learning-based deraining methods are supervisedly trained on synthetic rainy-clean pairs. The domain gap between the synthetic and real rains makes them less generalized to different real rainy scenes. Moreover, the existing methods mainly utilize the property of the two layers independently, while few of them have considered the mutually exclusive relationship between the two layers. In this work, we propose a novel non-local contrastive learning (NLCL) method for unsupervised image deraining. Consequently, we not only utilize the intrinsic self-similarity property within samples but also the mutually exclusive property between the two layers, so as to better differ the rain layer from the clean image. Specifically, the non-local self-similarity image layer patches as the positives are pulled together and similar rain layer patches as the negatives are pushed away. Thus the similar positive/negative samples that are close in the original space benefit us to enrich more discriminative representation. Apart from the self-similarity sampling strategy, we analyze how to choose an appropriate feature encoder in NLCL. Extensive experiments on different real rainy datasets demonstrate that the proposed method obtains state-of-the-art performance in real deraining.

Weicheng Xue, Baisong Xu, Kai Yang et al.

Modern AI accelerators provide high-throughput low-precision matrix engines, but their support for FP32 GEMM is often limited or inefficient. This work presents SGEMM-cube, a precision-recovery FP32 GEMM approximation on Ascend NPUs using FP16 Cube units. Rather than claiming bit-exact FP32 approximation, SGEMM-cube targets near-FP32 accuracy for inputs whose magnitudes are representable within the FP16 dynamic range. The method follows a two-component FP32-to-FP16 splitting strategy related to Ozaki-style and Ootomo-style schemes: each FP32 operand is represented by an FP16 high component and a scaled FP16 residual component, and the matrix product is reconstructed from the dominant high-high and high-low terms while omitting the low-low term. The main contribution of this paper is not a new splitting paradigm, but an architecture-specific realization and analysis of this precision-recovery scheme on Ascend NPUs. We analyze the effects of round-to-nearest conversion, underflow, residual scaling, and accumulation order under the Ascend execution model, and clarify the range and accuracy limitations of the approach. We further adapt standard high-performance GEMM techniques, including L1-aware blocking and double-buffered pipelining, to the software-managed memory hierarchy of Ascend NPUs. Experiments on Ascend 910A show that SGEMM-cube recovers substantially higher accuracy than native FP16 GEMM and approaches FP32 SGEMM accuracy for moderate-range inputs, while achieving up to 65.3 TFLOP/s, corresponding to 77\% of the FP32-equivalent peak defined by the three-GEMM decomposition cost. These results demonstrate that FP32-accuracy GEMM approximation can be made practical on FP16-only NPU matrix engines, provided that its range, error, and implementation constraints are explicitly managed.

Wangbo Yu, Jinbo Xing, Li Yuan et al.

Despite recent advancements in neural 3D reconstruction, the dependence on dense multi-view captures restricts their broader applicability. In this work, we propose \textbf{ViewCrafter}, a novel method for synthesizing high-fidelity novel views of generic scenes from single or sparse images with the prior of video diffusion model. Our method takes advantage of the powerful generation capabilities of video diffusion model and the coarse 3D clues offered by point-based representation to generate high-quality video frames with precise camera pose control. To further enlarge the generation range of novel views, we tailored an iterative view synthesis strategy together with a camera trajectory planning algorithm to progressively extend the 3D clues and the areas covered by the novel views. With ViewCrafter, we can facilitate various applications, such as immersive experiences with real-time rendering by efficiently optimizing a 3D-GS representation using the reconstructed 3D points and the generated novel views, and scene-level text-to-3D generation for more imaginative content creation. Extensive experiments on diverse datasets demonstrate the strong generalization capability and superior performance of our method in synthesizing high-fidelity and consistent novel views.

Xijie Xiang, Lin Zhu, Jianing Li et al.

The event camera is a novel bio-inspired vision sensor. When the brightness change exceeds the preset threshold, the sensor generates events asynchronously. The number of valid events directly affects the performance of event-based tasks, such as reconstruction, detection, and recognition. However, when in low-brightness or slow-moving scenes, events are often sparse and accompanied by noise, which poses challenges for event-based tasks. To solve these challenges, we propose an event temporal up-sampling algorithm1 to generate more effective and reliable events. The main idea of our algorithm is to generate up-sampling events on the event motion trajectory. First, we estimate the event motion trajectory by contrast maximization algorithm and then up-sampling the events by temporal point processes. Experimental results show that up-sampling events can provide more effective information and improve the performance of downstream tasks, such as improving the quality of reconstructed images and increasing the accuracy of object detection.

Yi Chang, Yun Guo, Yuntong Ye et al.

Most of the existing learning-based deraining methods are supervisedly trained on synthetic rainy-clean pairs. The domain gap between the synthetic and real rain makes them less generalized to complex real rainy scenes. Moreover, the existing methods mainly utilize the property of the image or rain layers independently, while few of them have considered their mutually exclusive relationship. To solve above dilemma, we explore the intrinsic intra-similarity within each layer and inter-exclusiveness between two layers and propose an unsupervised non-local contrastive learning (NLCL) deraining method. The non-local self-similarity image patches as the positives are tightly pulled together, rain patches as the negatives are remarkably pushed away, and vice versa. On one hand, the intrinsic self-similarity knowledge within positive/negative samples of each layer benefits us to discover more compact representation; on the other hand, the mutually exclusive property between the two layers enriches the discriminative decomposition. Thus, the internal self-similarity within each layer (similarity) and the external exclusive relationship of the two layers (dissimilarity) serving as a generic image prior jointly facilitate us to unsupervisedly differentiate the rain from clean image. We further discover that the intrinsic dimension of the non-local image patches is generally higher than that of the rain patches. This motivates us to design an asymmetric contrastive loss to precisely model the compactness discrepancy of the two layers for better discriminative decomposition. In addition, considering that the existing real rain datasets are of low quality, either small scale or downloaded from the internet, we collect a real large-scale dataset under various rainy kinds of weather that contains high-resolution rainy images.

Guangyao Chen, Peixi Peng, Guoqi Li et al.

Due to the binary spike signals making converting the traditional high-power multiply-accumulation (MAC) into a low-power accumulation (AC) available, the brain-inspired Spiking Neural Networks (SNNs) are gaining more and more attention. However, the binary spike propagation of the Full-Spike Neural Networks (FSNN) with limited time steps is prone to significant information loss. To improve performance, several state-of-the-art SNN models trained from scratch inevitably bring many non-spike operations. The non-spike operations cause additional computational consumption and may not be deployed on some neuromorphic hardware where only spike operation is allowed. To train a large-scale FSNN with high performance, this paper proposes a novel Dual-Stream Training (DST) method which adds a detachable Auxiliary Accumulation Pathway (AAP) to the full spiking residual networks. The accumulation in AAP could compensate for the information loss during the forward and backward of full spike propagation, and facilitate the training of the FSNN. In the test phase, the AAP could be removed and only the FSNN remained. This not only keeps the lower energy consumption but also makes our model easy to deploy. Moreover, for some cases where the non-spike operations are available, the APP could also be retained in test inference and improve feature discrimination by introducing a little non-spike consumption. Extensive experiments on ImageNet, DVS Gesture, and CIFAR10-DVS datasets demonstrate the effectiveness of DST.

Chenlin Zhou, Liutao Yu, Zhaokun Zhou et al.

Spiking neural networks (SNNs) offer a promising energy-efficient alternative to artificial neural networks, due to their event-driven spiking computation. However, some foundation SNN backbones (including Spikformer and SEW ResNet) suffer from non-spike computations (integer-float multiplications) caused by the structure of their residual connections. These non-spike computations increase SNNs' power consumption and make them unsuitable for deployment on mainstream neuromorphic hardware. In this paper, we analyze the spike-driven behavior of the residual connection methods in SNNs. We then present Spikingformer, a novel spiking transformer backbone that merges the MS Residual connection with Self-Attention in a biologically plausible way to address the non-spike computation challenge in Spikformer while maintaining global modeling capabilities. We evaluate Spikingformer across 13 datasets spanning large static images, neuromorphic data, and natural language tasks, and demonstrate the effectiveness and universality of Spikingformer, setting a vital benchmark for spiking neural networks. In addition, with the spike-driven features and global modeling capabilities, Spikingformer is expected to become a more efficient general-purpose SNN backbone towards energy-efficient artificial intelligence. Code: https://github.com/TheBrainLab/Spikingformer

Lantian Xue, Yixiong Zou, Peixi Peng et al.

Person Re-identification (Re-ID) has attracted great attention due to its promising real-world applications. However, in practice, it is always costly to annotate the training data to train a Re-ID model, and it still remains challenging to reduce the annotation cost while maintaining the performance for the Re-ID task. To solve this problem, we propose the Annotation Efficient Person Re-Identification method to select image pairs from an alternative pair set according to the fallibility and diversity of pairs, and train the Re-ID model based on the annotation. Specifically, we design an annotation and training framework to firstly reduce the size of the alternative pair set by clustering all images considering the locality of features, secondly select images pairs from intra-/inter-cluster samples for human to annotate, thirdly re-assign clusters according to the annotation, and finally train the model with the re-assigned clusters. During the pair selection, we seek for valuable pairs according to pairs' fallibility and diversity, which includes an intra-cluster criterion to construct image pairs with the most chaotic samples and the representative samples within clusters, an inter-cluster criterion to construct image pairs between clusters based on the second-order Wasserstein distance, and a diversity criterion for clusterbased pair selection. Combining all criteria above, a greedy strategy is developed to solve the pair selection problem. Finally, the above clustering-selecting-annotating-reassigning-training procedure will be repeated until the annotation budget is reached. Extensive experiments on three widely adopted Re-ID datasets show that we can greatly reduce the annotation cost while achieving better performance compared with state-of-the-art works.

Munan Ning, Donghuan Lu, Yujia Xie et al.

Unsupervised domain adaption has been widely adopted in tasks with scarce annotated data. Unfortunately, mapping the target-domain distribution to the source-domain unconditionally may distort the essential structural information of the target-domain data, leading to inferior performance. To address this issue, we firstly propose to introduce active sample selection to assist domain adaptation regarding the semantic segmentation task. By innovatively adopting multiple anchors instead of a single centroid, both source and target domains can be better characterized as multimodal distributions, in which way more complementary and informative samples are selected from the target domain. With only a little workload to manually annotate these active samples, the distortion of the target-domain distribution can be effectively alleviated, achieving a large performance gain. In addition, a powerful semi-supervised domain adaptation strategy is proposed to alleviate the long-tail distribution problem and further improve the segmentation performance. Extensive experiments are conducted on public datasets, and the results demonstrate that the proposed approach outperforms state-of-the-art methods by large margins and achieves similar performance to the fully-supervised upperbound, i.e., 71.4% mIoU on GTA5 and 71.8% mIoU on SYNTHIA. The effectiveness of each component is also verified by thorough ablation studies.

Wangbo Yu, Li Yuan, Yan-Pei Cao et al.

Recent advances in diffusion models have enabled 3D generation from a single image. However, current methods often produce suboptimal results for novel views, with blurred textures and deviations from the reference image, limiting their practical applications. In this paper, we introduce HiFi-123, a method designed for high-fidelity and multi-view consistent 3D generation. Our contributions are twofold: First, we propose a Reference-Guided Novel View Enhancement (RGNV) technique that significantly improves the fidelity of diffusion-based zero-shot novel view synthesis methods. Second, capitalizing on the RGNV, we present a novel Reference-Guided State Distillation (RGSD) loss. When incorporated into the optimization-based image-to-3D pipeline, our method significantly improves 3D generation quality, achieving state-of-the-art performance. Comprehensive evaluations demonstrate the effectiveness of our approach over existing methods, both qualitatively and quantitatively. Video results are available on the project page.

Bin Lin, Yunyang Ge, Xinhua Cheng et al.

We introduce Open-Sora Plan, an open-source project that aims to contribute a large generation model for generating desired high-resolution videos with long durations based on various user inputs. Our project comprises multiple components for the entire video generation process, including a Wavelet-Flow Variational Autoencoder, a Joint Image-Video Skiparse Denoiser, and various condition controllers. Moreover, many assistant strategies for efficient training and inference are designed, and a multi-dimensional data curation pipeline is proposed for obtaining desired high-quality data. Benefiting from efficient thoughts, our Open-Sora Plan achieves impressive video generation results in both qualitative and quantitative evaluations. We hope our careful design and practical experience can inspire the video generation research community. All our codes and model weights are publicly available at \url{https://github.com/PKU-YuanGroup/Open-Sora-Plan}.

Haiyang Zhou, Xinhua Cheng, Wangbo Yu et al.

3D scene generation is in high demand across various domains, including virtual reality, gaming, and the film industry. Owing to the powerful generative capabilities of text-to-image diffusion models that provide reliable priors, the creation of 3D scenes using only text prompts has become viable, thereby significantly advancing researches in text-driven 3D scene generation. In order to obtain multiple-view supervision from 2D diffusion models, prevailing methods typically employ the diffusion model to generate an initial local image, followed by iteratively outpainting the local image using diffusion models to gradually generate scenes. Nevertheless, these outpainting-based approaches prone to produce global inconsistent scene generation results without high degree of completeness, restricting their broader applications. To tackle these problems, we introduce HoloDreamer, a framework that first generates high-definition panorama as a holistic initialization of the full 3D scene, then leverage 3D Gaussian Splatting (3D-GS) to quickly reconstruct the 3D scene, thereby facilitating the creation of view-consistent and fully enclosed 3D scenes. Specifically, we propose Stylized Equirectangular Panorama Generation, a pipeline that combines multiple diffusion models to enable stylized and detailed equirectangular panorama generation from complex text prompts. Subsequently, Enhanced Two-Stage Panorama Reconstruction is introduced, conducting a two-stage optimization of 3D-GS to inpaint the missing region and enhance the integrity of the scene. Comprehensive experiments demonstrated that our method outperforms prior works in terms of overall visual consistency and harmony as well as reconstruction quality and rendering robustness when generating fully enclosed scenes.

Man Yao, Jiakui Hu, Tianxiang Hu et al.

Neuromorphic computing, which exploits Spiking Neural Networks (SNNs) on neuromorphic chips, is a promising energy-efficient alternative to traditional AI. CNN-based SNNs are the current mainstream of neuromorphic computing. By contrast, no neuromorphic chips are designed especially for Transformer-based SNNs, which have just emerged, and their performance is only on par with CNN-based SNNs, offering no distinct advantage. In this work, we propose a general Transformer-based SNN architecture, termed as ``Meta-SpikeFormer", whose goals are: 1) Lower-power, supports the spike-driven paradigm that there is only sparse addition in the network; 2) Versatility, handles various vision tasks; 3) High-performance, shows overwhelming performance advantages over CNN-based SNNs; 4) Meta-architecture, provides inspiration for future next-generation Transformer-based neuromorphic chip designs. Specifically, we extend the Spike-driven Transformer in \citet{yao2023spike} into a meta architecture, and explore the impact of structure, spike-driven self-attention, and skip connection on its performance. On ImageNet-1K, Meta-SpikeFormer achieves 80.0\% top-1 accuracy (55M), surpassing the current state-of-the-art (SOTA) SNN baselines (66M) by 3.7\%. This is the first direct training SNN backbone that can simultaneously supports classification, detection, and segmentation, obtaining SOTA results in SNNs. Finally, we discuss the inspiration of the meta SNN architecture for neuromorphic chip design. Source code and models are available at \url{https://github.com/BICLab/Spike-Driven-Transformer-V2}.

Xiao Wang, Shiao Wang, Yuhe Ding et al.

In the post-deep learning era, the Transformer architecture has demonstrated its powerful performance across pre-trained big models and various downstream tasks. However, the enormous computational demands of this architecture have deterred many researchers. To further reduce the complexity of attention models, numerous efforts have been made to design more efficient methods. Among them, the State Space Model (SSM), as a possible replacement for the self-attention based Transformer model, has drawn more and more attention in recent years. In this paper, we give the first comprehensive review of these works and also provide experimental comparisons and analysis to better demonstrate the features and advantages of SSM. Specifically, we first give a detailed description of principles to help the readers quickly capture the key ideas of SSM. After that, we dive into the reviews of existing SSMs and their various applications, including natural language processing, computer vision, graph, multi-modal and multi-media, point cloud/event stream, time series data, and other domains. In addition, we give statistical comparisons and analysis of these models and hope it helps the readers to understand the effectiveness of different structures on various tasks. Then, we propose possible research points in this direction to better promote the development of the theoretical model and application of SSM. More related works will be continuously updated on the following GitHub: https://github.com/Event-AHU/Mamba_State_Space_Model_Paper_List.

Hanyu Zhou, Yi Chang, Zhiwei Shi et al.

Optical flow has made great progress in clean scenes, while suffers degradation under adverse weather due to the violation of the brightness constancy and gradient continuity assumptions of optical flow. Typically, existing methods mainly adopt domain adaptation to transfer motion knowledge from clean to degraded domain through one-stage adaptation. However, this direct adaptation is ineffective, since there exists a large gap due to adverse weather and scene style between clean and real degraded domains. Moreover, even within the degraded domain itself, static weather (e.g., fog) and dynamic weather (e.g., rain) have different impacts on optical flow. To address above issues, we explore synthetic degraded domain as an intermediate bridge between clean and real degraded domains, and propose a cumulative homogeneous-heterogeneous adaptation framework for real adverse weather optical flow. Specifically, for clean-degraded transfer, our key insight is that static weather possesses the depth-association homogeneous feature which does not change the intrinsic motion of the scene, while dynamic weather additionally introduces the heterogeneous feature which results in a significant boundary discrepancy in warp errors between clean and degraded domains. For synthetic-real transfer, we figure out that cost volume correlation shares a similar statistical histogram between synthetic and real degraded domains, benefiting to holistically aligning the homogeneous correlation distribution for synthetic-real knowledge distillation. Under this unified framework, the proposed method can progressively and explicitly transfer knowledge from clean scenes to real adverse weather. In addition, we further collect a real adverse weather dataset with manually annotated optical flow labels and perform extensive experiments to verify the superiority of the proposed method.

Chenlin Zhou, Han Zhang, Zhaokun Zhou et al.

Spiking Transformers, which integrate Spiking Neural Networks (SNNs) with Transformer architectures, have attracted significant attention due to their potential for energy efficiency and high performance. However, existing models in this domain still suffer from suboptimal performance. We introduce several innovations to improve the performance: i) We propose a novel spike-form Q-K attention mechanism, tailored for SNNs, which efficiently models the importance of token or channel dimensions through binary vectors with linear complexity. ii) We incorporate the hierarchical structure, which significantly benefits the performance of both the brain and artificial neural networks, into spiking transformers to obtain multi-scale spiking representation. iii) We design a versatile and powerful patch embedding module with a deformed shortcut specifically for spiking transformers. Together, we develop QKFormer, a hierarchical spiking transformer based on Q-K attention with direct training. QKFormer shows significantly superior performance over existing state-of-the-art SNN models on various mainstream datasets. Notably, with comparable size to Spikformer (66.34 M, 74.81%), QKFormer (64.96 M) achieves a groundbreaking top-1 accuracy of 85.65% on ImageNet-1k, substantially outperforming Spikformer by 10.84%. To our best knowledge, this is the first time that directly training SNNs have exceeded 85% accuracy on ImageNet-1K. The code and models are publicly available at https://github.com/zhouchenlin2096/QKFormer

Mingjian Ni, Guangyao Chen, Xiawu Zheng et al.

The sparsity of Deep Neural Networks is well investigated to maximize the performance and reduce the size of overparameterized networks as possible. Existing methods focus on pruning parameters in the training process by using thresholds and metrics. Meanwhile, feature similarity between different layers has not been discussed sufficiently before, which could be rigorously proved to be highly correlated to the network sparsity in this paper. Inspired by interlayer feature similarity in overparameterized models, we investigate the intrinsic link between network sparsity and interlayer feature similarity. Specifically, we prove that reducing interlayer feature similarity based on Centered Kernel Alignment (CKA) improves the sparsity of the network by using information bottleneck theory. Applying such theory, we propose a plug-and-play CKA-based Sparsity Regularization for sparse network training, dubbed CKA-SR, which utilizes CKA to reduce feature similarity between layers and increase network sparsity. In other words, layers of our sparse network tend to have their own identity compared to each other. Experimentally, we plug the proposed CKA-SR into the training process of sparse network training methods and find that CKA-SR consistently improves the performance of several State-Of-The-Art sparse training methods, especially at extremely high sparsity. Code is included in the supplementary materials.

Youneng Bao, Fangyang Meng, Wen Tan et al.

In this paper, a unified transformation method in learned image compression(LIC) is proposed from the perspective of modulation. Firstly, the quantization in LIC is considered as a generalized channel with additive uniform noise. Moreover, the LIC is interpreted as a particular communication system according to the consistency in structures and optimization objectives. Thus, the technology of communication systems can be applied to guide the design of modules in LIC. Furthermore, a unified transform method based on signal modulation (TSM) is defined. In the view of TSM, the existing transformation methods are mathematically reduced to a linear modulation. A series of transformation methods, e.g. TPM and TJM, are obtained by extending to nonlinear modulation. The experimental results on various datasets and backbone architectures verify that the effectiveness and robustness of the proposed method. More importantly, it further confirms the feasibility of guiding LIC design from a communication perspective. For example, when backbone architecture is hyperprior combining context model, our method achieves 3.52$\%$ BD-rate reduction over GDN on Kodak dataset without increasing complexity.

Xiao Wang, Yu Jin, Wentao Wu et al.

Object detection in event streams has emerged as a cutting-edge research area, demonstrating superior performance in low-light conditions, scenarios with motion blur, and rapid movements. Current detectors leverage spiking neural networks, Transformers, or convolutional neural networks as their core architectures, each with its own set of limitations including restricted performance, high computational overhead, or limited local receptive fields. This paper introduces a novel MoE (Mixture of Experts) heat conduction-based object detection algorithm that strikingly balances accuracy and computational efficiency. Initially, we employ a stem network for event data embedding, followed by processing through our innovative MoE-HCO blocks. Each block integrates various expert modules to mimic heat conduction within event streams. Subsequently, an IoU-based query selection module is utilized for efficient token extraction, which is then channeled into a detection head for the final object detection process. Furthermore, we are pleased to introduce EvDET200K, a novel benchmark dataset for event-based object detection. Captured with a high-definition Prophesee EVK4-HD event camera, this dataset encompasses 10 distinct categories, 200,000 bounding boxes, and 10,054 samples, each spanning 2 to 5 seconds. We also provide comprehensive results from over 15 state-of-the-art detectors, offering a solid foundation for future research and comparison. The source code of this paper will be released on: https://github.com/Event-AHU/OpenEvDET

Yabin Zhu, Xiao Wang, Chenglong Li et al.

Existing datasets for RGB-DVS tracking are collected with DVS346 camera and their resolution ($346 \times 260$) is low for practical applications. Actually, only visible cameras are deployed in many practical systems, and the newly designed neuromorphic cameras may have different resolutions. The latest neuromorphic sensors can output high-definition event streams, but it is very difficult to achieve strict alignment between events and frames on both spatial and temporal views. Therefore, how to achieve accurate tracking with unaligned neuromorphic and visible sensors is a valuable but unresearched problem. In this work, we formally propose the task of object tracking using unaligned neuromorphic and visible cameras. We build the first unaligned frame-event dataset CRSOT collected with a specially built data acquisition system, which contains 1,030 high-definition RGB-Event video pairs, 304,974 video frames. In addition, we propose a novel unaligned object tracking framework that can realize robust tracking even using the loosely aligned RGB-Event data. Specifically, we extract the template and search regions of RGB and Event data and feed them into a unified ViT backbone for feature embedding. Then, we propose uncertainty perception modules to encode the RGB and Event features, respectively, then, we propose a modality uncertainty fusion module to aggregate the two modalities. These three branches are jointly optimized in the training phase. Extensive experiments demonstrate that our tracker can collaborate the dual modalities for high-performance tracking even without strictly temporal and spatial alignment. The source code, dataset, and pre-trained models will be released at https://github.com/Event-AHU/Cross_Resolution_SOT.

Xiao Wang, Qian Zhu, Jiandong Jin et al.

Existing pedestrian attribute recognition (PAR) algorithms are mainly developed based on a static image, however, the performance is unreliable in challenging scenarios, such as heavy occlusion, motion blur, etc. In this work, we propose to understand human attributes using video frames that can fully use temporal information by fine-tuning a pre-trained multi-modal foundation model efficiently. Specifically, we formulate the video-based PAR as a vision-language fusion problem and adopt a pre-trained foundation model CLIP to extract the visual features. More importantly, we propose a novel spatiotemporal side-tuning strategy to achieve parameter-efficient optimization of the pre-trained vision foundation model. To better utilize the semantic information, we take the full attribute list that needs to be recognized as another input and transform the attribute words/phrases into the corresponding sentence via split, expand, and prompt operations. Then, the text encoder of CLIP is utilized for embedding processed attribute descriptions. The averaged visual tokens and text tokens are concatenated and fed into a fusion Transformer for multi-modal interactive learning. The enhanced tokens will be fed into a classification head for pedestrian attribute prediction. Extensive experiments on two large-scale video-based PAR datasets fully validated the effectiveness of our proposed framework. The source code of this paper is available at https://github.com/Event-AHU/OpenPAR.

Hongwei Yi, Shitong Shao, Tian Ye et al.

In this technical report, we present Magic 1-For-1 (Magic141), an efficient video generation model with optimized memory consumption and inference latency. The key idea is simple: factorize the text-to-video generation task into two separate easier tasks for diffusion step distillation, namely text-to-image generation and image-to-video generation. We verify that with the same optimization algorithm, the image-to-video task is indeed easier to converge over the text-to-video task. We also explore a bag of optimization tricks to reduce the computational cost of training the image-to-video (I2V) models from three aspects: 1) model convergence speedup by using a multi-modal prior condition injection; 2) inference latency speed up by applying an adversarial step distillation, and 3) inference memory cost optimization with parameter sparsification. With those techniques, we are able to generate 5-second video clips within 3 seconds. By applying a test time sliding window, we are able to generate a minute-long video within one minute with significantly improved visual quality and motion dynamics, spending less than 1 second for generating 1 second video clips on average. We conduct a series of preliminary explorations to find out the optimal tradeoff between computational cost and video quality during diffusion step distillation and hope this could be a good foundation model for open-source explorations. The code and the model weights are available at https://github.com/DA-Group-PKU/Magic-1-For-1.

Xiao Wang, Fuling Wang, Haowen Wang et al.

X-ray image based medical report generation achieves significant progress in recent years with the help of the large language model, however, these models have not fully exploited the effective information in visual image regions, resulting in reports that are linguistically sound but insufficient in describing key diseases. In this paper, we propose a novel associative memory-enhanced X-ray report generation model that effectively mimics the process of professional doctors writing medical reports. It considers both the mining of global and local visual information and associates historical report information to better complete the writing of the current report. Specifically, given an X-ray image, we first utilize a classification model along with its activation maps to accomplish the mining of visual regions highly associated with diseases and the learning of disease query tokens. Then, we employ a visual Hopfield network to establish memory associations for disease-related tokens, and a report Hopfield network to retrieve report memory information. This process facilitates the generation of high-quality reports based on a large language model and achieves state-of-the-art performance on multiple benchmark datasets, including the IU X-ray, MIMIC-CXR, and Chexpert Plus. The source code of this work is released on \url{https://github.com/Event-AHU/Medical_Image_Analysis}.

Dianze Li, Jianing Li, Xu Liu et al.

Combining the complementary benefits of frames and events has been widely used for object detection in challenging scenarios. However, most object detection methods use two independent Artificial Neural Network (ANN) branches, limiting cross-modality information interaction across the two visual streams and encountering challenges in extracting temporal cues from event streams with low power consumption. To address these challenges, we propose HDI-Former, a Hybrid Dynamic Interaction ANN-SNN Transformer, marking the first trial to design a directly trained hybrid ANN-SNN architecture for high-accuracy and energy-efficient object detection using frames and events. Technically, we first present a novel semantic-enhanced self-attention mechanism that strengthens the correlation between image encoding tokens within the ANN Transformer branch for better performance. Then, we design a Spiking Swin Transformer branch to model temporal cues from event streams with low power consumption. Finally, we propose a bio-inspired dynamic interaction mechanism between ANN and SNN sub-networks for cross-modality information interaction. The results demonstrate that our HDI-Former outperforms eleven state-of-the-art methods and our four baselines by a large margin. Our SNN branch also shows comparable performance to the ANN with the same architecture while consuming 10.57$\times$ less energy on the DSEC-Detection dataset. Our open-source code is available in the supplementary material.

Xiao Wang, Yao Rong, Shiao Wang et al.

Pattern recognition based on RGB-Event data is a newly arising research topic and previous works usually learn their features using CNN or Transformer. As we know, CNN captures the local features well and the cascaded self-attention mechanisms are good at extracting the long-range global relations. It is intuitive to combine them for high-performance RGB-Event based video recognition, however, existing works fail to achieve a good balance between the accuracy and model parameters, as shown in Fig.~\ref{firstimage}. In this work, we propose a novel RGB-Event based recognition framework termed TSCFormer, which is a relatively lightweight CNN-Transformer model. Specifically, we mainly adopt the CNN as the backbone network to first encode both RGB and Event data. Meanwhile, we initialize global tokens as the input and fuse them with RGB and Event features using the BridgeFormer module. It captures the global long-range relations well between both modalities and maintains the simplicity of the whole model architecture at the same time. The enhanced features will be projected and fused into the RGB and Event CNN blocks, respectively, in an interactive manner using F2E and F2V modules. Similar operations are conducted for other CNN blocks to achieve adaptive fusion and local-global feature enhancement under different resolutions. Finally, we concatenate these three features and feed them into the classification head for pattern recognition. Extensive experiments on two large-scale RGB-Event benchmark datasets (PokerEvent and HARDVS) fully validated the effectiveness of our proposed TSCFormer. The source code and pre-trained models will be released at https://github.com/Event-AHU/TSCFormer.