Songyuan Li

Xintian Wu, Qihang Zhang, Yiming Wu et al.

Formulated as a conditional generation problem, face animation aims at synthesizing continuous face images from a single source image driven by a set of conditional face motion. Previous works mainly model the face motion as conditions with 1D or 2D representation (e.g., action units, emotion codes, landmark), which often leads to low-quality results in some complicated scenarios such as continuous generation and largepose transformation. To tackle this problem, the conditions are supposed to meet two requirements, i.e., motion information preserving and geometric continuity. To this end, we propose a novel representation based on a 3D geometric flow, termed facial flow, to represent the natural motion of the human face at any pose. Compared with other previous conditions, the proposed facial flow well controls the continuous changes to the face. After that, in order to utilize the facial flow for face editing, we build a synthesis framework generating continuous images with conditional facial flows. To fully take advantage of the motion information of facial flows, a hierarchical conditional framework is designed to combine the extracted multi-scale appearance features from images and motion features from flows in a hierarchical manner. The framework then decodes multiple fused features back to images progressively. Experimental results demonstrate the effectiveness of our method compared to other state-of-the-art methods.

Yifeng Chen, Guangchen Lin, Songyuan Li et al.

Panoptic segmentation aims to perform instance segmentation for foreground instances and semantic segmentation for background stuff simultaneously. The typical top-down pipeline concentrates on two key issues: 1) how to effectively model the intrinsic interaction between semantic segmentation and instance segmentation, and 2) how to properly handle occlusion for panoptic segmentation. Intuitively, the complementarity between semantic segmentation and instance segmentation can be leveraged to improve the performance. Besides, we notice that using detection/mask scores is insufficient for resolving the occlusion problem. Motivated by these observations, we propose a novel deep panoptic segmentation scheme based on a bidirectional learning pipeline. Moreover, we introduce a plug-and-play occlusion handling algorithm to deal with the occlusion between different object instances. The experimental results on COCO panoptic benchmark validate the effectiveness of our proposed method. Codes will be released soon at https://github.com/Mooonside/BANet.

Junyi Feng, Songyuan Li, Xi Li et al.

Real-time semantic video segmentation is a challenging task due to the strict requirements of inference speed. Recent approaches mainly devote great efforts to reducing the model size for high efficiency. In this paper, we rethink this problem from a different viewpoint: using knowledge contained in compressed videos. We propose a simple and effective framework, dubbed TapLab, to tap into resources from the compressed domain. Specifically, we design a fast feature warping module using motion vectors for acceleration. To reduce the noise introduced by motion vectors, we design a residual-guided correction module and a residual-guided frame selection module using residuals. TapLab significantly reduces redundant computations of the state-of-the-art fast semantic image segmentation models, running 3 to 10 times faster with controllable accuracy degradation. The experimental results show that TapLab achieves 70.6% mIoU on the Cityscapes dataset at 99.8 FPS with a single GPU card for the 1024x2048 videos. A high-speed version even reaches the speed of 160+ FPS. Codes will be available soon at https://github.com/Sixkplus/TapLab.

Gelu Liu, Teng Wang, Zhijie Wu et al.

Autonomous bicycles offer a promising agile solution for urban mobility and last-mile logistics, however, conventional control strategies often struggle with their underactuated nonlinear dynamics, suffering from sensitivity to model mismatches and limited adaptability to real-world uncertainties. To address this, this paper presents CycleRL, the first sim-to-real deep reinforcement learning framework designed for robust autonomous bicycle control. Our approach trains an end-to-end neural control policy within the high-fidelity NVIDIA Isaac Sim environment, leveraging Proximal Policy Optimization (PPO) to circumvent the need for an explicit dynamics model. The framework features a composite reward function tailored for concurrent balance maintenance, velocity tracking, and steering control. Crucially, systematic domain randomization is employed to bridge the simulation-to-reality gap and facilitate direct transfer. In simulation, CycleRL achieves considerable performance, including a 99.90% balance success rate, a low steering tracking error of 1.15°, and a velocity tracking error of 0.18 m/s. These quantitative results, coupled with successful hardware transfer, validate DRL as an effective paradigm for autonomous bicycle control, offering superior adaptability over traditional methods. Video demonstrations are available at https://anony6f05.github.io/CycleRL/.

Ziyang Gong, Wenhao Li, Oliver Ma et al.

Multimodal Large Language Models (MLLMs) have achieved remarkable progress in various multimodal tasks. To pursue higher intelligence in space, MLLMs require integrating multiple spatial capabilities, even for handling simple and normal tasks. However, existing benchmarks struggle to comprehensively evaluate the spatial intelligence of common MLLMs from the atomic level to the compositional level. To fill this gap, we present SpaCE-10, a comprehensive benchmark for compositional spatial evaluations. In SpaCE-10, we define 10 atomic spatial capabilities, which are combined to form 8 compositional capabilities. Based on these definitions, we propose a novel hierarchical annotation pipeline to generate high-quality and diverse question-answer (QA) pairs. With over 150+ hours of human expert effort, we obtain over 5k QA pairs for 811 real indoor scenes in SpaCE-10, which covers various evaluation settings like point cloud input and multi-choice QA. We conduct an extensive evaluation of common MLLMs on SpaCE-10 and find that even the most advanced MLLM still lags behind humans by large margins. Through our careful study, we also draw several significant findings that benefit the MLLM community. For example, we reveal that the shortcoming of counting capability greatly limits the compositional spatial capabilities of existing MLLMs.

Songyuan Li, Jia Hu, Geyong Min et al.

Foundation models (FMs) such as GPT-4 exhibit exceptional generative capabilities across diverse downstream tasks through fine-tuning. Split Federated Learning (SFL) facilitates privacy-preserving FM fine-tuning on resource-constrained local devices by offloading partial FM computations to edge servers, enabling device-edge synergistic fine-tuning. Practical edge networks often host multiple SFL tenants to support diversified downstream tasks. However, existing research primarily focuses on single-tenant SFL scenarios, and lacks tailored incentive mechanisms for multi-tenant settings, which are essential to effectively coordinate self-interested local devices for participation in various downstream tasks, ensuring that each SFL tenant's distinct FM fine-tuning requirements (e.g., FM types, performance targets, and fine-tuning deadlines) are met. To address this gap, we propose a novel Price-Incentive Mechanism (PRINCE) that guides multiple SFL tenants to offer strategic price incentives, which solicit high-quality device participation for efficient FM fine-tuning. Specifically, we first develop a bias-resilient global SFL model aggregation scheme to eliminate model biases caused by independent device participation. We then derive a rigorous SFL convergence bound to evaluate the contributions of heterogeneous devices to FM performance improvements, guiding the incentive strategies of SFL tenants. Furthermore, we model inter-tenant device competition as a congestion game for Stackelberg equilibrium (SE) analysis, deriving each SFL tenant's optimal incentive strategy. Extensive simulations involving four representative SFL tenant types (ViT, BERT, Whisper, and LLaMA) across diverse data modalities (text, images, and audio) demonstrate that PRINCE accelerates FM fine-tuning by up to 3.07x compared to state-of-the-art approaches, while consistently meeting fine-tuning performance targets.

Songyuan Li, Jia Hu, Ahmed M. Abdelmoniem et al.

Recent Mixture-of-Experts (MoE)-based large language models (LLMs) such as Qwen-MoE and DeepSeek-MoE are transforming generative AI in natural language processing. However, these models require vast and diverse training data. Federated learning (FL) addresses this challenge by leveraging private data from heterogeneous edge devices for privacy-preserving MoE training. Nonetheless, traditional FL approaches require devices to host local MoE models, which is impractical for resource-constrained devices due to large model sizes. To address this, we propose DeepFusion, the first scalable federated MoE training framework that enables the fusion of heterogeneous on-device LLM knowledge via federated knowledge distillation, yielding a knowledge-abundant global MoE model. Specifically, DeepFusion features each device to independently configure and train an on-device LLM tailored to its own needs and hardware limitations. Furthermore, we propose a novel View-Aligned Attention (VAA) module that integrates multi-stage feature representations from the global MoE model to construct a predictive perspective aligned with on-device LLMs, thereby enabling effective cross-architecture knowledge distillation. By explicitly aligning predictive perspectives, VAA resolves the view-mismatch problem in traditional federated knowledge distillation, which arises from heterogeneity in model architectures and prediction behaviors between on-device LLMs and the global MoE model. Experiments with industry-level MoE models (Qwen-MoE and DeepSeek-MoE) and real-world datasets (medical and finance) demonstrate that DeepFusion achieves performance close to centralized MoE training. Compared with key federated MoE baselines, DeepFusion reduces communication costs by up to 71% and improves token perplexity by up to 5.28%.

Songyuan Li, Teng Wang, Jinrong Tang et al.

Achieving fully analog neural computation requires hardware that can natively implement both linear and nonlinear operations with high efficiency. While analogue matrix-vector multiplication has advanced via compute-in-memory architectures, nonlinear activation functions remain a bottleneck, often requiring digital or hybrid solutions. Inspired by the Kolmogorov-Arnold framework, we propose KANalogue, a fully analogue implementation of Kolmogorov-Arnold Networks (KANs) using negative differential resistance devices as physical realizations of learnable univariate basis functions. By leveraging the intrinsic negative differential resistance characteristics of tunnel diodes fabricated from NbSi2N4/HfSi2N4 heterostructures, we construct coordinate-wise nonlinearities with distinct curvature and support profiles. We extract I-V data from fabricated armchair and zigzag devices, fit high-order polynomials to emulate diode behavior in software, and train KANs on vision benchmarks using these learned basis functions. Our results demonstrate that KANalogue can approximate complex functions with minimal parameters while maintaining classification accuracy competitive with digital baselines. This work bridges device-level physics and function approximation theory, charting a path toward scalable, energy-efficient analogue machine learning systems.

Songyuan Li, Jia Hu, Geyong Min et al.

The convergence of edge computing and AI gives rise to Edge-AI, which enables the deployment of real-time AI applications and services at the network edge. One of the fundamental research issues in Edge-AI is edge inference acceleration, which aims to realize low-latency high-accuracy DNN inference services by leveraging the fine-grained offloading of partitioned inference tasks from end devices to edge servers. However, existing research has yet to adopt a practical Edge-AI market perspective, which would systematically explore the personalized inference needs of AI users (e.g., inference accuracy, latency, and task complexity), the revenue incentives for AI service providers that offer edge inference services, and multi-stakeholder governance within a market-oriented context. To bridge this gap, we propose an Auction-based Edge Inference Pricing Mechanism (AERIA) for revenue maximization to tackle the multi-dimensional optimization problem of DNN model partition, edge inference pricing, and resource allocation. We investigate the multi-exit device-edge synergistic inference scheme for on-demand DNN inference acceleration, and analyse the auction dynamics amongst the AI service providers, AI users and edge infrastructure provider. Owing to the strategic mechanism design via randomized consensus estimate and cost sharing techniques, the Edge-AI market attains several desirable properties, including competitiveness in revenue maximization, incentive compatibility, and envy-freeness, which are crucial to maintain the effectiveness, truthfulness, and fairness of our auction outcomes. The extensive simulation experiments based on four representative DNN inference workloads demonstrate that our AERIA mechanism significantly outperforms several state-of-the-art approaches in revenue maximization, demonstrating the efficacy of AERIA for on-demand DNN inference in the Edge-AI market.

Jiabao Cui, Pengyi Zhang, Songyuan Li et al.

High-capacity image steganography, aimed at concealing a secret image in a cover image, is a technique to preserve sensitive data, e.g., faces and fingerprints. Previous methods focus on the security during transmission and subsequently run a risk of privacy leakage after the restoration of secret images at the receiving end. To address this issue, we propose a framework, called Multitask Identity-Aware Image Steganography (MIAIS), to achieve direct recognition on container images without restoring secret images. The key issue of the direct recognition is to preserve identity information of secret images into container images and make container images look similar to cover images at the same time. Thus, we introduce a simple content loss to preserve the identity information, and design a minimax optimization to deal with the contradictory aspects. We demonstrate that the robustness results can be transferred across different cover datasets. In order to be flexible for the secret image restoration in some cases, we incorporate an optional restoration network into our method, providing a multitask framework. The experiments under the multitask scenario show the effectiveness of our framework compared with other visual information hiding methods and state-of-the-art high-capacity image steganography methods.

Pengyi Zhang, Huanzhang Dou, Wenhu Zhang et al.

Gait recognition has a rapid development in recent years. However, gait recognition in the wild is not well explored yet. An obvious reason could be ascribed to the lack of diverse training data from the perspective of intrinsic and extrinsic factors. To remedy this problem, we propose to construct a large-scale gait dataset with the help of controllable computer simulation. In detail, to diversify the intrinsic factors of gait, we generate numerous characters with diverse attributes and empower them with various types of walking styles. To diversify the extrinsic factors of gait, we build a complicated scene with a dense camera layout. Finally, we design an automated generation toolkit under Unity3D for simulating the walking scenario and capturing the gait data automatically. As a result, we obtain an in-the-wild gait dataset, called VersatileGait, which has more than one million silhouette sequences of 10,000 subjects with diverse scenarios. VersatileGait possesses several nice properties, including huge dataset size, diverse pedestrian attributes, complicated camera layout, high-quality annotations, small domain gap with the real one, good scalability for new demands, and no privacy issues. Based on VersatileGait, we propose series of experiments and applications for both research exploration of gait in the wild and practical applications. Our dataset and its corresponding generation toolkit will be publicly available for further studies.

Jabeen Summaira, Xi Li, Amin Muhammad Shoib et al.

Deep Learning has implemented a wide range of applications and has become increasingly popular in recent years. The goal of multimodal deep learning is to create models that can process and link information using various modalities. Despite the extensive development made for unimodal learning, it still cannot cover all the aspects of human learning. Multimodal learning helps to understand and analyze better when various senses are engaged in the processing of information. This paper focuses on multiple types of modalities, i.e., image, video, text, audio, body gestures, facial expressions, and physiological signals. Detailed analysis of past and current baseline approaches and an in-depth study of recent advancements in multimodal deep learning applications has been provided. A fine-grained taxonomy of various multimodal deep learning applications is proposed, elaborating on different applications in more depth. Architectures and datasets used in these applications are also discussed, along with their evaluation metrics. Last, main issues are highlighted separately for each domain along with their possible future research directions.

Peng Sun, Wenhu Zhang, Huanyu Wang et al.

RGB-D salient object detection (SOD) is usually formulated as a problem of classification or regression over two modalities, i.e., RGB and depth. Hence, effective RGBD feature modeling and multi-modal feature fusion both play a vital role in RGB-D SOD. In this paper, we propose a depth-sensitive RGB feature modeling scheme using the depth-wise geometric prior of salient objects. In principle, the feature modeling scheme is carried out in a depth-sensitive attention module, which leads to the RGB feature enhancement as well as the background distraction reduction by capturing the depth geometry prior. Moreover, to perform effective multi-modal feature fusion, we further present an automatic architecture search approach for RGB-D SOD, which does well in finding out a feasible architecture from our specially designed multi-modal multi-scale search space. Extensive experiments on seven standard benchmarks demonstrate the effectiveness of the proposed approach against the state-of-the-art.

Hui Wang, Jian Tian, Songyuan Li et al.

Unsupervised domain adaptation (UDA) typically carries out knowledge transfer from a label-rich source domain to an unlabeled target domain by adversarial learning. In principle, existing UDA approaches mainly focus on the global distribution alignment between domains while ignoring the intrinsic local distribution properties. Motivated by this observation, we propose an end-to-end structure-conditioned adversarial learning scheme (SCAL) that is able to preserve the intra-class compactness during domain distribution alignment. By using local structures as structure-aware conditions, the proposed scheme is implemented in a structure-conditioned adversarial learning pipeline. The above learning procedure is iteratively performed by alternating between local structures establishment and structure-conditioned adversarial learning. Experimental results demonstrate the effectiveness of the proposed scheme in UDA scenarios.

Huanzhang Dou, Wenhu Zhang, Pengyi Zhang et al.

With the motivation of practical gait recognition applications, we propose to automatically create a large-scale synthetic gait dataset (called VersatileGait) by a game engine, which consists of around one million silhouette sequences of 11,000 subjects with fine-grained attributes in various complicated scenarios. Compared with existing real gait datasets with limited samples and simple scenarios, the proposed VersatileGait dataset possesses several nice properties, including huge dataset size, high sample diversity, high-quality annotations, multi-pitch angles, small domain gap with the real one, etc. Furthermore, we investigate the effectiveness of our dataset (e.g., domain transfer after pretraining). Then, we use the fine-grained attributes from VersatileGait to promote gait recognition in both accuracy and speed, and meanwhile justify the gait recognition performance under multi-pitch angle settings. Additionally, we explore a variety of potential applications for research.Extensive experiments demonstrate the value and effective-ness of the proposed VersatileGait in gait recognition along with its associated applications. We will release both VersatileGait and its corresponding data generation toolkit for further studies.

Xuewei Li, Songyuan Li, Bourahla Omar et al.

Knowledge distillation, aimed at transferring the knowledge from a heavy teacher network to a lightweight student network, has emerged as a promising technique for compressing neural networks. However, due to the capacity gap between the heavy teacher and the lightweight student, there still exists a significant performance gap between them. In this paper, we see knowledge distillation in a fresh light, using the knowledge gap, or the residual, between a teacher and a student as guidance to train a much more lightweight student, called a res-student. We combine the student and the res-student into a new student, where the res-student rectifies the errors of the former student. Such a residual-guided process can be repeated until the user strikes the balance between accuracy and cost. At inference time, we propose a sample-adaptive strategy to decide which res-students are not necessary for each sample, which can save computational cost. Experimental results show that we achieve competitive performance with 18.04$\%$, 23.14$\%$, 53.59$\%$, and 56.86$\%$ of the teachers' computational costs on the CIFAR-10, CIFAR-100, Tiny-ImageNet, and ImageNet datasets. Finally, we do thorough theoretical and empirical analysis for our method.

Huanyu Wang, Zequn Qin, Songyuan Li et al.

Dynamic routing networks, aimed at finding the best routing paths in the networks, have achieved significant improvements to neural networks in terms of accuracy and efficiency. In this paper, we see dynamic routing networks in a fresh light, formulating a routing method as a mapping from a sample space to a routing space. From the perspective of space mapping, prevalent methods of dynamic routing didn't consider how inference paths would be distributed in the routing space. Thus, we propose a novel method, termed CoDiNet, to model the relationship between a sample space and a routing space by regularizing the distribution of routing paths with the properties of consistency and diversity. Specifically, samples with similar semantics should be mapped into the same area in routing space, while those with dissimilar semantics should be mapped into different areas. Moreover, we design a customizable dynamic routing module, which can strike a balance between accuracy and efficiency. When deployed upon ResNet models, our method achieves higher performance and effectively reduces average computational cost on four widely used datasets.

Songyuan Li, Junyi Feng, Xi Li

Recent approaches for fast semantic video segmentation have reduced redundancy by warping feature maps across adjacent frames, greatly speeding up the inference phase. However, the accuracy drops seriously owing to the errors incurred by warping. In this paper, we propose a novel framework and design a simple and effective correction stage after warping. Specifically, we build a non-key-frame CNN, fusing warped context features with current spatial details. Based on the feature fusion, our Context Feature Rectification~(CFR) module learns the model's difference from a per-frame model to correct the warped features. Furthermore, our Residual-Guided Attention~(RGA) module utilizes the residual maps in the compressed domain to help CRF focus on error-prone regions. Results on Cityscapes show that the accuracy significantly increases from $67.3\%$ to $71.6\%$, and the speed edges down from $65.5$ FPS to $61.8$ FPS at a resolution of $1024\times 2048$. For non-rigid categories, e.g., ``human'' and ``object'', the improvements are even higher than 18 percentage points.

Peng Sun, Jiaxiang Wu, Songyuan Li et al.

To satisfy the stringent requirements on computational resources in the field of real-time semantic segmentation, most approaches focus on the hand-crafted design of light-weight segmentation networks. Recently, Neural Architecture Search (NAS) has been used to search for the optimal building blocks of networks automatically, but the network depth, downsampling strategy, and feature aggregation way are still set in advance by trial and error. In this paper, we propose a joint search framework, called AutoRTNet, to automate the design of these strategies. Specifically, we propose hyper-cells to jointly decide the network depth and downsampling strategy, and an aggregation cell to achieve automatic multi-scale feature aggregation. Experimental results show that AutoRTNet achieves 73.9% mIoU on the Cityscapes test set and 110.0 FPS on an NVIDIA TitanXP GPU card with 768x1536 input images.