Zhongdao Wang

Xin Liu, Zhongdao Wang, Yali Li et al.

A mainstream type of current self-supervised learning methods pursues a general-purpose representation that can be well transferred to downstream tasks, typically by optimizing on a given pretext task such as instance discrimination. In this work, we argue that existing pretext tasks inevitably introduce biases into the learned representation, which in turn leads to biased transfer performance on various downstream tasks. To cope with this issue, we propose Maximum Entropy Coding (MEC), a more principled objective that explicitly optimizes on the structure of the representation, so that the learned representation is less biased and thus generalizes better to unseen downstream tasks. Inspired by the principle of maximum entropy in information theory, we hypothesize that a generalizable representation should be the one that admits the maximum entropy among all plausible representations. To make the objective end-to-end trainable, we propose to leverage the minimal coding length in lossy data coding as a computationally tractable surrogate for the entropy, and further derive a scalable reformulation of the objective that allows fast computation. Extensive experiments demonstrate that MEC learns a more generalizable representation than previous methods based on specific pretext tasks. It achieves state-of-the-art performance consistently on various downstream tasks, including not only ImageNet linear probe, but also semi-supervised classification, object detection, instance segmentation, and object tracking. Interestingly, we show that existing batch-wise and feature-wise self-supervised objectives could be seen equivalent to low-order approximations of MEC. Code and pre-trained models are available at https://github.com/xinliu20/MEC.

Zhaopeng Dou, Zhongdao Wang, Yali Li et al.

This paper aims to learn a domain-generalizable (DG) person re-identification (ReID) representation from large-scale videos \textbf{without any annotation}. Prior DG ReID methods employ limited labeled data for training due to the high cost of annotation, which restricts further advances. To overcome the barriers of data and annotation, we propose to utilize large-scale unsupervised data for training. The key issue lies in how to mine identity information. To this end, we propose an Identity-seeking Self-supervised Representation learning (ISR) method. ISR constructs positive pairs from inter-frame images by modeling the instance association as a maximum-weight bipartite matching problem. A reliability-guided contrastive loss is further presented to suppress the adverse impact of noisy positive pairs, ensuring that reliable positive pairs dominate the learning process. The training cost of ISR scales approximately linearly with the data size, making it feasible to utilize large-scale data for training. The learned representation exhibits superior generalization ability. \textbf{Without human annotation and fine-tuning, ISR achieves 87.0\% Rank-1 on Market-1501 and 56.4\% Rank-1 on MSMT17}, outperforming the best supervised domain-generalizable method by 5.0\% and 19.5\%, respectively. In the pre-training$\rightarrow$fine-tuning scenario, ISR achieves state-of-the-art performance, with 88.4\% Rank-1 on MSMT17. The code is at \url{https://github.com/dcp15/ISR_ICCV2023_Oral}.

Junsong Chen, Jincheng Yu, Chongjian Ge et al.

The most advanced text-to-image (T2I) models require significant training costs (e.g., millions of GPU hours), seriously hindering the fundamental innovation for the AIGC community while increasing CO2 emissions. This paper introduces PIXART-$α$, a Transformer-based T2I diffusion model whose image generation quality is competitive with state-of-the-art image generators (e.g., Imagen, SDXL, and even Midjourney), reaching near-commercial application standards. Additionally, it supports high-resolution image synthesis up to 1024px resolution with low training cost, as shown in Figure 1 and 2. To achieve this goal, three core designs are proposed: (1) Training strategy decomposition: We devise three distinct training steps that separately optimize pixel dependency, text-image alignment, and image aesthetic quality; (2) Efficient T2I Transformer: We incorporate cross-attention modules into Diffusion Transformer (DiT) to inject text conditions and streamline the computation-intensive class-condition branch; (3) High-informative data: We emphasize the significance of concept density in text-image pairs and leverage a large Vision-Language model to auto-label dense pseudo-captions to assist text-image alignment learning. As a result, PIXART-$α$'s training speed markedly surpasses existing large-scale T2I models, e.g., PIXART-$α$ only takes 10.8% of Stable Diffusion v1.5's training time (675 vs. 6,250 A100 GPU days), saving nearly \$300,000 (\$26,000 vs. \$320,000) and reducing 90% CO2 emissions. Moreover, compared with a larger SOTA model, RAPHAEL, our training cost is merely 1%. Extensive experiments demonstrate that PIXART-$α$ excels in image quality, artistry, and semantic control. We hope PIXART-$α$ will provide new insights to the AIGC community and startups to accelerate building their own high-quality yet low-cost generative models from scratch.

Chongjian Ge, Junsong Chen, Enze Xie et al.

Perception systems in modern autonomous driving vehicles typically take inputs from complementary multi-modal sensors, e.g., LiDAR and cameras. However, in real-world applications, sensor corruptions and failures lead to inferior performances, thus compromising autonomous safety. In this paper, we propose a robust framework, called MetaBEV, to address extreme real-world environments involving overall six sensor corruptions and two extreme sensor-missing situations. In MetaBEV, signals from multiple sensors are first processed by modal-specific encoders. Subsequently, a set of dense BEV queries are initialized, termed meta-BEV. These queries are then processed iteratively by a BEV-Evolving decoder, which selectively aggregates deep features from either LiDAR, cameras, or both modalities. The updated BEV representations are further leveraged for multiple 3D prediction tasks. Additionally, we introduce a new M2oE structure to alleviate the performance drop on distinct tasks in multi-task joint learning. Finally, MetaBEV is evaluated on the nuScenes dataset with 3D object detection and BEV map segmentation tasks. Experiments show MetaBEV outperforms prior arts by a large margin on both full and corrupted modalities. For instance, when the LiDAR signal is missing, MetaBEV improves 35.5% detection NDS and 17.7% segmentation mIoU upon the vanilla BEVFusion model; and when the camera signal is absent, MetaBEV still achieves 69.2% NDS and 53.7% mIoU, which is even higher than previous works that perform on full-modalities. Moreover, MetaBEV performs fairly against previous methods in both canonical perception and multi-task learning settings, refreshing state-of-the-art nuScenes BEV map segmentation with 70.4% mIoU.

Qinpeng Cui, Yixuan Liu, Xinyi Zhang et al.

Diffusion-based image super-resolution (SR) models have attracted substantial interest due to their powerful image restoration capabilities. However, prevailing diffusion models often struggle to strike an optimal balance between efficiency and performance. Typically, they either neglect to exploit the potential of existing extensive pretrained models, limiting their generative capacity, or they necessitate a dozens of forward passes starting from random noises, compromising inference efficiency. In this paper, we present DoSSR, a Domain Shift diffusion-based SR model that capitalizes on the generative powers of pretrained diffusion models while significantly enhancing efficiency by initiating the diffusion process with low-resolution (LR) images. At the core of our approach is a domain shift equation that integrates seamlessly with existing diffusion models. This integration not only improves the use of diffusion prior but also boosts inference efficiency. Moreover, we advance our method by transitioning the discrete shift process to a continuous formulation, termed as DoS-SDEs. This advancement leads to the fast and customized solvers that further enhance sampling efficiency. Empirical results demonstrate that our proposed method achieves state-of-the-art performance on synthetic and real-world datasets, while notably requiring only 5 sampling steps. Compared to previous diffusion prior based methods, our approach achieves a remarkable speedup of 5-7 times, demonstrating its superior efficiency. Code: https://github.com/QinpengCui/DoSSR.

Zhaopeng Dou, Zhongdao Wang, Weihua Chen et al.

Current person image retrieval methods have achieved great improvements in accuracy metrics. However, they rarely describe the reliability of the prediction. In this paper, we propose an Uncertainty-Aware Learning (UAL) method to remedy this issue. UAL aims at providing reliability-aware predictions by considering data uncertainty and model uncertainty simultaneously. Data uncertainty captures the ``noise" inherent in the sample, while model uncertainty depicts the model's confidence in the sample's prediction. Specifically, in UAL, (1) we propose a sampling-free data uncertainty learning method to adaptively assign weights to different samples during training, down-weighting the low-quality ambiguous samples. (2) we leverage the Bayesian framework to model the model uncertainty by assuming the parameters of the network follow a Bernoulli distribution. (3) the data uncertainty and the model uncertainty are jointly learned in a unified network, and they serve as two fundamental criteria for the reliability assessment: if a probe is high-quality (low data uncertainty) and the model is confident in the prediction of the probe (low model uncertainty), the final ranking will be assessed as reliable. Experiments under the risk-controlled settings and the multi-query settings show the proposed reliability assessment is effective. Our method also shows superior performance on three challenging benchmarks under the vanilla single query settings.

Jianhao Li, Tianyu Sun, Zhongdao Wang et al.

This paper proposes an algorithm for automatically labeling 3D objects from 2D point or box prompts, especially focusing on applications in autonomous driving. Unlike previous arts, our auto-labeler predicts 3D shapes instead of bounding boxes and does not require training on a specific dataset. We propose a Segment, Lift, and Fit (SLF) paradigm to achieve this goal. Firstly, we segment high-quality instance masks from the prompts using the Segment Anything Model (SAM) and transform the remaining problem into predicting 3D shapes from given 2D masks. Due to the ill-posed nature of this problem, it presents a significant challenge as multiple 3D shapes can project into an identical mask. To tackle this issue, we then lift 2D masks to 3D forms and employ gradient descent to adjust their poses and shapes until the projections fit the masks and the surfaces conform to surrounding LiDAR points. Notably, since we do not train on a specific dataset, the SLF auto-labeler does not overfit to biased annotation patterns in the training set as other methods do. Thus, the generalization ability across different datasets improves. Experimental results on the KITTI dataset demonstrate that the SLF auto-labeler produces high-quality bounding box annotations, achieving an AP@0.5 IoU of nearly 90\%. Detectors trained with the generated pseudo-labels perform nearly as well as those trained with actual ground-truth annotations. Furthermore, the SLF auto-labeler shows promising results in detailed shape predictions, providing a potential alternative for the occupancy annotation of dynamic objects.

Zhongdao Wang, Zhaopeng Dou, Jingwei Zhang et al.

In this paper, we are interested in learning a generalizable person re-identification (re-ID) representation from unlabeled videos. Compared with 1) the popular unsupervised re-ID setting where the training and test sets are typically under the same domain, and 2) the popular domain generalization (DG) re-ID setting where the training samples are labeled, our novel scenario combines their key challenges: the training samples are unlabeled, and collected form various domains which do no align with the test domain. In other words, we aim to learn a representation in an unsupervised manner and directly use the learned representation for re-ID in novel domains. To fulfill this goal, we make two main contributions: First, we propose Cycle Association (CycAs), a scalable self-supervised learning method for re-ID with low training complexity; and second, we construct a large-scale unlabeled re-ID dataset named LMP-video, tailored for the proposed method. Specifically, CycAs learns re-ID features by enforcing cycle consistency of instance association between temporally successive video frame pairs, and the training cost is merely linear to the data size, making large-scale training possible. On the other hand, the LMP-video dataset is extremely large, containing 50 million unlabeled person images cropped from over 10K Youtube videos, therefore is sufficient to serve as fertile soil for self-supervised learning. Trained on LMP-video, we show that CycAs learns good generalization towards novel domains. The achieved results sometimes even outperform supervised domain generalizable models. Remarkably, CycAs achieves 82.2% Rank-1 on Market-1501 and 49.0% Rank-1 on MSMT17 with zero human annotation, surpassing state-of-the-art supervised DG re-ID methods. Moreover, we also demonstrate the superiority of CycAs under the canonical unsupervised re-ID and the pretrain-and-finetune scenarios.

Jilai Zheng, Pin Tang, Zhongdao Wang et al.

Perceiving the world as 3D occupancy supports embodied agents to avoid collision with any types of obstacle. While open-vocabulary image understanding has prospered recently, how to bind the predicted 3D occupancy grids with open-world semantics still remains under-explored due to limited open-world annotations. Hence, instead of building our model from scratch, we try to blend 2D foundation models, specifically a depth model MiDaS and a semantic model CLIP, to lift the semantics to 3D space, thus fulfilling 3D occupancy. However, building upon these foundation models is not trivial. First, the MiDaS faces the depth ambiguity problem, i.e., it only produces relative depth but fails to estimate bin depth for feature lifting. Second, the CLIP image features lack high-resolution pixel-level information, which limits the 3D occupancy accuracy. Third, open vocabulary is often trapped by the long-tail problem. To address these issues, we propose VEON for Vocabulary-Enhanced Occupancy predictioN by not only assembling but also adapting these foundation models. We first equip MiDaS with a Zoedepth head and low-rank adaptation (LoRA) for relative-metric-bin depth transformation while reserving beneficial depth prior. Then, a lightweight side adaptor network is attached to the CLIP vision encoder to generate high-resolution features for fine-grained 3D occupancy prediction. Moreover, we design a class reweighting strategy to give priority to the tail classes. With only 46M trainable parameters and zero manual semantic labels, VEON achieves 15.14 mIoU on Occ3D-nuScenes, and shows the capability of recognizing objects with open-vocabulary categories, meaning that our VEON is label-efficient, parameter-efficient, and precise enough.

Song Wang, Zhongdao Wang, Jiawei Yu et al.

Vision-centric semantic occupancy prediction plays a crucial role in autonomous driving, which requires accurate and reliable predictions from low-cost sensors. Although having notably narrowed the accuracy gap with LiDAR, there is still few research effort to explore the reliability in predicting semantic occupancy from camera. In this paper, we conduct a comprehensive evaluation of existing semantic occupancy prediction models from a reliability perspective for the first time. Despite the gradual alignment of camera-based models with LiDAR in term of accuracy, a significant reliability gap persists. To addresses this concern, we propose ReliOcc, a method designed to enhance the reliability of camera-based occupancy networks. ReliOcc provides a plug-and-play scheme for existing models, which integrates hybrid uncertainty from individual voxels with sampling-based noise and relative voxels through mix-up learning. Besides, an uncertainty-aware calibration strategy is devised to further enhance model reliability in offline mode. Extensive experiments under various settings demonstrate that ReliOcc significantly enhances model reliability while maintaining the accuracy of both geometric and semantic predictions. Importantly, our proposed approach exhibits robustness to sensor failures and out of domain noises during inference.

Mingkang Zhu, Xi Chen, Zhongdao Wang et al.

Recent advances in text-to-image model customization have underscored the importance of integrating new concepts with a few examples. Yet, these progresses are largely confined to widely recognized subjects, which can be learned with relative ease through models' adequate shared prior knowledge. In contrast, logos, characterized by unique patterns and textual elements, are hard to establish shared knowledge within diffusion models, thus presenting a unique challenge. To bridge this gap, we introduce the task of logo insertion. Our goal is to insert logo identities into diffusion models and enable their seamless synthesis in varied contexts. We present a novel two-phase pipeline LogoSticker to tackle this task. First, we propose the actor-critic relation pre-training algorithm, which addresses the nontrivial gaps in models' understanding of the potential spatial positioning of logos and interactions with other objects. Second, we propose a decoupled identity learning algorithm, which enables precise localization and identity extraction of logos. LogoSticker can generate logos accurately and harmoniously in diverse contexts. We comprehensively validate the effectiveness of LogoSticker over customization methods and large models such as DALLE~3. \href{https://mingkangz.github.io/logosticker}{Project page}.

Mingkang Zhu, Xi Chen, Zhongdao Wang et al.

Recent advancements in reinforcement learning from human feedback have shown that utilizing fine-grained token-level reward models can substantially enhance the performance of Proximal Policy Optimization (PPO) in aligning large language models. However, it is challenging to leverage such token-level reward as guidance for Direct Preference Optimization (DPO), since DPO is formulated as a sequence-level bandit problem. To address this challenge, this work decomposes the sequence-level PPO into a sequence of token-level proximal policy optimization problems and then frames the problem of token-level PPO with token-level reward guidance, from which closed-form optimal token-level policy and the corresponding token-level reward can be derived. Using the obtained reward and Bradley-Terry model, this work establishes a framework of computable loss functions with token-level reward guidance for DPO, and proposes a practical reward guidance based on the induced DPO reward. This formulation enables different tokens to exhibit varying degrees of deviation from reference policy based on their respective rewards. Experiment results demonstrate that our method achieves substantial performance improvements over DPO, with win rate gains of up to 7.5 points on MT-Bench, 6.2 points on AlpacaEval 2, and 4.3 points on Arena-Hard. Code is available at https://github.com/dvlab-research/TGDPO.

Junsong Chen, Chongjian Ge, Enze Xie et al.

In this paper, we introduce PixArt-Σ, a Diffusion Transformer model~(DiT) capable of directly generating images at 4K resolution. PixArt-Σrepresents a significant advancement over its predecessor, PixArt-α, offering images of markedly higher fidelity and improved alignment with text prompts. A key feature of PixArt-Σis its training efficiency. Leveraging the foundational pre-training of PixArt-α, it evolves from the `weaker' baseline to a `stronger' model via incorporating higher quality data, a process we term "weak-to-strong training". The advancements in PixArt-Σare twofold: (1) High-Quality Training Data: PixArt-Σincorporates superior-quality image data, paired with more precise and detailed image captions. (2) Efficient Token Compression: we propose a novel attention module within the DiT framework that compresses both keys and values, significantly improving efficiency and facilitating ultra-high-resolution image generation. Thanks to these improvements, PixArt-Σachieves superior image quality and user prompt adherence capabilities with significantly smaller model size (0.6B parameters) than existing text-to-image diffusion models, such as SDXL (2.6B parameters) and SD Cascade (5.1B parameters). Moreover, PixArt-Σ's capability to generate 4K images supports the creation of high-resolution posters and wallpapers, efficiently bolstering the production of high-quality visual content in industries such as film and gaming.

Xiangxuan Ren, Zhongdao Wang, Liping Hou et al.

Multimodal large language models (MLLMs) have made significant advancements in vision understanding and reasoning. However, the autoregressive Transformer architecture used by MLLMs requries tokenization on input images, which limits their ability to accurately ground objects within the 2D image space. This raises an important question: how can sequential language tokens be improved to better ground objects in 2D spatial space for MLLMs? To address this, we present a spatial representation method for grounding objects, namely GETok, that integrates a specialized vocabulary of learnable tokens into MLLMs. GETok first uses grid tokens to partition the image plane into structured spatial anchors, and then exploits offset tokens to enable precise and iterative refinement of localization predictions. By embedding spatial relationships directly into tokens, GETok significantly advances MLLMs in native 2D space reasoning without modifying the autoregressive architecture. Extensive experiments demonstrate that GETok achieves superior performance over the state-of-the-art methods across various referring tasks in both supervised fine-tuning and reinforcement learning settings.

Xiangxuan Ren, Zhongdao Wang, Pin Tang et al.

3D object detection is fundamental for safe and robust intelligent transportation systems. Current multi-modal 3D object detectors often rely on complex architectures and training strategies to achieve higher detection accuracy. However, these methods heavily rely on the LiDAR sensor so that they suffer from large performance drops when LiDAR is absent, which compromises the robustness and safety of autonomous systems in practical scenarios. Moreover, existing multi-modal detectors face difficulties in deployment on diverse hardware platforms, such as NPUs and FPGAs, due to their reliance on 3D sparse convolution operators, which are primarily optimized for NVIDIA GPUs. To address these challenges, we reconsider the role of LiDAR in the camera-LiDAR fusion paradigm and introduce a novel multi-modal 3D detector, LiteFusion. Instead of treating LiDAR point clouds as an independent modality with a separate feature extraction backbone, LiteFusion utilizes LiDAR data as a complementary source of geometric information to enhance camera-based detection. This straightforward approach completely eliminates the reliance on a 3D backbone, making the method highly deployment-friendly. Specifically, LiteFusion integrates complementary features from LiDAR points into image features within a quaternion space, where the orthogonal constraints are well-preserved during network training. This helps model domain-specific relations across modalities, yielding a compact cross-modal embedding. Experiments on the nuScenes dataset show that LiteFusion improves the baseline vision-based detector by +20.4% mAP and +19.7% NDS with a minimal increase in parameters (1.1%) without using dedicated LiDAR encoders. Notably, even in the absence of LiDAR input, LiteFusion maintains strong results , highlighting its favorable robustness and effectiveness across diverse fusion paradigms and deployment scenarios.

Zhongdao Wang, Liang Zheng, Yixuan Liu et al.

Modern multiple object tracking (MOT) systems usually follow the \emph{tracking-by-detection} paradigm. It has 1) a detection model for target localization and 2) an appearance embedding model for data association. Having the two models separately executed might lead to efficiency problems, as the running time is simply a sum of the two steps without investigating potential structures that can be shared between them. Existing research efforts on real-time MOT usually focus on the association step, so they are essentially real-time association methods but not real-time MOT system. In this paper, we propose an MOT system that allows target detection and appearance embedding to be learned in a shared model. Specifically, we incorporate the appearance embedding model into a single-shot detector, such that the model can simultaneously output detections and the corresponding embeddings. We further propose a simple and fast association method that works in conjunction with the joint model. In both components the computation cost is significantly reduced compared with former MOT systems, resulting in a neat and fast baseline for future follow-ups on real-time MOT algorithm design. To our knowledge, this work reports the first (near) real-time MOT system, with a running speed of 22 to 40 FPS depending on the input resolution. Meanwhile, its tracking accuracy is comparable to the state-of-the-art trackers embodying separate detection and embedding (SDE) learning ($64.4\%$ MOTA \vs $66.1\%$ MOTA on MOT-16 challenge). Code and models are available at \url{https://github.com/Zhongdao/Towards-Realtime-MOT}.

Pin Tang, Zhongdao Wang, Guoqing Wang et al.

Vision-based perception for autonomous driving requires an explicit modeling of a 3D space, where 2D latent representations are mapped and subsequent 3D operators are applied. However, operating on dense latent spaces introduces a cubic time and space complexity, which limits scalability in terms of perception range or spatial resolution. Existing approaches compress the dense representation using projections like Bird's Eye View (BEV) or Tri-Perspective View (TPV). Although efficient, these projections result in information loss, especially for tasks like semantic occupancy prediction. To address this, we propose SparseOcc, an efficient occupancy network inspired by sparse point cloud processing. It utilizes a lossless sparse latent representation with three key innovations. Firstly, a 3D sparse diffuser performs latent completion using spatially decomposed 3D sparse convolutional kernels. Secondly, a feature pyramid and sparse interpolation enhance scales with information from others. Finally, the transformer head is redesigned as a sparse variant. SparseOcc achieves a remarkable 74.9% reduction on FLOPs over the dense baseline. Interestingly, it also improves accuracy, from 12.8% to 14.1% mIOU, which in part can be attributed to the sparse representation's ability to avoid hallucinations on empty voxels.

Guoqing Wang, Zhongdao Wang, Pin Tang et al.

Existing solutions for 3D semantic occupancy prediction typically treat the task as a one-shot 3D voxel-wise segmentation perception problem. These discriminative methods focus on learning the mapping between the inputs and occupancy map in a single step, lacking the ability to gradually refine the occupancy map and the reasonable scene imaginative capacity to complete the local regions somewhere. In this paper, we introduce OccGen, a simple yet powerful generative perception model for the task of 3D semantic occupancy prediction. OccGen adopts a ''noise-to-occupancy'' generative paradigm, progressively inferring and refining the occupancy map by predicting and eliminating noise originating from a random Gaussian distribution. OccGen consists of two main components: a conditional encoder that is capable of processing multi-modal inputs, and a progressive refinement decoder that applies diffusion denoising using the multi-modal features as conditions. A key insight of this generative pipeline is that the diffusion denoising process is naturally able to model the coarse-to-fine refinement of the dense 3D occupancy map, therefore producing more detailed predictions. Extensive experiments on several occupancy benchmarks demonstrate the effectiveness of the proposed method compared to the state-of-the-art methods. For instance, OccGen relatively enhances the mIoU by 9.5%, 6.3%, and 13.3% on nuScenes-Occupancy dataset under the muli-modal, LiDAR-only, and camera-only settings, respectively. Moreover, as a generative perception model, OccGen exhibits desirable properties that discriminative models cannot achieve, such as providing uncertainty estimates alongside its multiple-step predictions.

Yunlong Lin, Zixu Lin, Kunjie Lin et al.

Photo retouching has become integral to contemporary visual storytelling, enabling users to capture aesthetics and express creativity. While professional tools such as Adobe Lightroom offer powerful capabilities, they demand substantial expertise and manual effort. In contrast, existing AI-based solutions provide automation but often suffer from limited adjustability and poor generalization, failing to meet diverse and personalized editing needs. To bridge this gap, we introduce JarvisArt, a multi-modal large language model (MLLM)-driven agent that understands user intent, mimics the reasoning process of professional artists, and intelligently coordinates over 200 retouching tools within Lightroom. JarvisArt undergoes a two-stage training process: an initial Chain-of-Thought supervised fine-tuning to establish basic reasoning and tool-use skills, followed by Group Relative Policy Optimization for Retouching (GRPO-R) to further enhance its decision-making and tool proficiency. We also propose the Agent-to-Lightroom Protocol to facilitate seamless integration with Lightroom. To evaluate performance, we develop MMArt-Bench, a novel benchmark constructed from real-world user edits. JarvisArt demonstrates user-friendly interaction, superior generalization, and fine-grained control over both global and local adjustments, paving a new avenue for intelligent photo retouching. Notably, it outperforms GPT-4o with a 60% improvement in average pixel-level metrics on MMArt-Bench for content fidelity, while maintaining comparable instruction-following capabilities. Project Page: https://jarvisart.vercel.app/.

Jingjing Ren, Wenbo Li, Zhongdao Wang et al.

Demand for 2K video synthesis is rising with increasing consumer expectations for ultra-clear visuals. While diffusion transformers (DiTs) have demonstrated remarkable capabilities in high-quality video generation, scaling them to 2K resolution remains computationally prohibitive due to quadratic growth in memory and processing costs. In this work, we propose Turbo2K, an efficient and practical framework for generating detail-rich 2K videos while significantly improving training and inference efficiency. First, Turbo2K operates in a highly compressed latent space, reducing computational complexity and memory footprint, making high-resolution video synthesis feasible. However, the high compression ratio of the VAE and limited model size impose constraints on generative quality. To mitigate this, we introduce a knowledge distillation strategy that enables a smaller student model to inherit the generative capacity of a larger, more powerful teacher model. Our analysis reveals that, despite differences in latent spaces and architectures, DiTs exhibit structural similarities in their internal representations, facilitating effective knowledge transfer. Second, we design a hierarchical two-stage synthesis framework that first generates multi-level feature at lower resolutions before guiding high-resolution video generation. This approach ensures structural coherence and fine-grained detail refinement while eliminating redundant encoding-decoding overhead, further enhancing computational efficiency.Turbo2K achieves state-of-the-art efficiency, generating 5-second, 24fps, 2K videos with significantly reduced computational cost. Compared to existing methods, Turbo2K is up to 20$\times$ faster for inference, making high-resolution video generation more scalable and practical for real-world applications.

Xiaomin Li, Yixuan Liu, Takashi Isobe et al.

In text-to-image (T2I) generation applications, negative embeddings have proven to be a simple yet effective approach for enhancing generation quality. Typically, these negative embeddings are derived from user-defined negative prompts, which, while being functional, are not necessarily optimal. In this paper, we introduce ReNeg, an end-to-end method designed to learn improved Negative embeddings guided by a Reward model. We employ a reward feedback learning framework and integrate classifier-free guidance (CFG) into the training process, which was previously utilized only during inference, thus enabling the effective learning of negative embeddings. We also propose two strategies for learning both global and per-sample negative embeddings. Extensive experiments show that the learned negative embedding significantly outperforms null-text and handcrafted counterparts, achieving substantial improvements in human preference alignment. Additionally, the negative embedding learned within the same text embedding space exhibits strong generalization capabilities. For example, using the same CLIP text encoder, the negative embedding learned on SD1.5 can be seamlessly transferred to text-to-image or even text-to-video models such as ControlNet, ZeroScope, and VideoCrafter2, resulting in consistent performance improvements across the board.

Zhongdao Wang, Guodongfang Zhao, Jingjing Ren et al.

Diffusion-based generative models have demonstrated exceptional promise in the video super-resolution (VSR) task, achieving a substantial advancement in detail generation relative to prior methods. However, these approaches face significant computational efficiency challenges. For instance, current techniques may require tens of minutes to super-resolve a mere 2-second, 1080p video. In this paper, we present TurboVSR, an ultra-efficient diffusion-based video super-resolution model. Our core design comprises three key aspects: (1) We employ an autoencoder with a high compression ratio of 32$\times$32$\times$8 to reduce the number of tokens. (2) Highly compressed latents pose substantial challenges for training. We introduce factorized conditioning to mitigate the learning complexity: we first learn to super-resolve the initial frame; subsequently, we condition the super-resolution of the remaining frames on the high-resolution initial frame and the low-resolution subsequent frames. (3) We convert the pre-trained diffusion model to a shortcut model to enable fewer sampling steps, further accelerating inference. As a result, TurboVSR performs on par with state-of-the-art VSR methods, while being 100+ times faster, taking only 7 seconds to process a 2-second long 1080p video. TurboVSR also supports image resolution by considering image as a one-frame video. Our efficient design makes SR beyond 1080p possible, results on 4K (3648$\times$2048) image SR show surprising fine details.

Mingkang Zhu, Xi Chen, Zhongdao Wang et al.

Recent diffusion model customization has shown impressive results in incorporating subject or style concepts with a handful of images. However, the modular composition of multiple concepts into a customized model, aimed to efficiently merge decentralized-trained concepts without influencing their identities, remains unresolved. Modular customization is essential for applications like concept stylization and multi-concept customization using concepts trained by different users. Existing post-training methods are only confined to a fixed set of concepts, and any different combinations require a new round of retraining. In contrast, instant merging methods often cause identity loss and interference of individual merged concepts and are usually limited to a small number of concepts. To address these issues, we propose BlockLoRA, an instant merging method designed to efficiently combine multiple concepts while accurately preserving individual concepts' identity. With a careful analysis of the underlying reason for interference, we develop the Randomized Output Erasure technique to minimize the interference of different customized models. Additionally, Blockwise LoRA Parameterization is proposed to reduce the identity loss during instant model merging. Extensive experiments validate the effectiveness of BlockLoRA, which can instantly merge 15 concepts of people, subjects, scenes, and styles with high fidelity.

Mingkang Zhu, Xi Chen, Zhongdao Wang et al.

As Large language models (LLMs) are increasingly deployed in diverse applications, faithfully integrating evolving factual knowledge into these models remains a critical challenge. Continued pre-training on paraphrased data has shown empirical promise for enhancing knowledge acquisition. However, this approach is often costly and unreliable, as it relies on external models or manual effort for rewriting, and may inadvertently alter the factual content. In this work, we hypothesize and empirically show that an LLM's ability to continually predict the same factual knowledge tokens given diverse paraphrased contexts is positively correlated with its capacity to extract that knowledge via question-answering. Based on this view and aiming to improve generalization to diverse paraphrased contexts, we introduce two strategies to enhance LLMs' ability to predict the same knowledge tokens given varied contexts, thereby enhancing knowledge acquisition. First, we propose formatting-based data augmentation, which diversifies documents conveying the same knowledge by altering document formats rather than their content, thereby preserving factual integrity. Second, we adopt sharpness-aware minimization as the optimizer to better improve generalization. Extensive experiments demonstrate our methods' effectiveness in both continued pre-training and instruction tuning, and further gains can be achieved by combining with paraphrased data.

Zhi-Lin Huang, Yixuan Liu, Chujun Qin et al.

Recent advancements in diffusion models have significantly facilitated text-guided video editing. However, there is a relative scarcity of research on image-guided video editing, a method that empowers users to edit videos by merely indicating a target object in the initial frame and providing an RGB image as reference, without relying on the text prompts. In this paper, we propose a novel Image-guided Video Editing Diffusion model, termed IVEDiff for the image-guided video editing. IVEDiff is built on top of image editing models, and is equipped with learnable motion modules to maintain the temporal consistency of edited video. Inspired by self-supervised learning concepts, we introduce a masked motion modeling fine-tuning strategy that empowers the motion module's capabilities for capturing inter-frame motion dynamics, while preserving the capabilities for intra-frame semantic correlations modeling of the base image editing model. Moreover, an optical-flow-guided motion reference network is proposed to ensure the accurate propagation of information between edited video frames, alleviating the misleading effects of invalid information. We also construct a benchmark to facilitate further research. The comprehensive experiments demonstrate that our method is able to generate temporally smooth edited videos while robustly dealing with various editing objects with high quality.

Tianyu Sun, Jianhao Li, Xueqian Zhang et al.

This paper studies point cloud perception within outdoor environments. Existing methods face limitations in recognizing objects located at a distance or occluded, due to the sparse nature of outdoor point clouds. In this work, we observe a significant mitigation of this problem by accumulating multiple temporally consecutive point cloud sweeps, resulting in a remarkable improvement in perception accuracy. However, the computation cost also increases, hindering previous approaches from utilizing a large number of point cloud sweeps. To tackle this challenge, we find that a considerable portion of points in the accumulated point cloud is redundant, and discarding these points has minimal impact on perception accuracy. We introduce a simple yet effective Gumbel Spatial Pruning (GSP) layer that dynamically prunes points based on a learned end-to-end sampling. The GSP layer is decoupled from other network components and thus can be seamlessly integrated into existing point cloud network architectures. Without incurring additional computational overhead, we increase the number of point cloud sweeps from 10, a common practice, to as many as 40. Consequently, there is a significant enhancement in perception performance. For instance, in nuScenes 3D object detection and BEV map segmentation tasks, our pruning strategy improves several 3D perception baseline methods.

Zhenyu Wang, Enze Xie, Aoxue Li et al.

Despite significant advancements in text-to-image models for generating high-quality images, these methods still struggle to ensure the controllability of text prompts over images in the context of complex text prompts, especially when it comes to retaining object attributes and relationships. In this paper, we propose CompAgent, a training-free approach for compositional text-to-image generation, with a large language model (LLM) agent as its core. The fundamental idea underlying CompAgent is premised on a divide-and-conquer methodology. Given a complex text prompt containing multiple concepts including objects, attributes, and relationships, the LLM agent initially decomposes it, which entails the extraction of individual objects, their associated attributes, and the prediction of a coherent scene layout. These individual objects can then be independently conquered. Subsequently, the agent performs reasoning by analyzing the text, plans and employs the tools to compose these isolated objects. The verification and human feedback mechanism is finally incorporated into our agent to further correct the potential attribute errors and refine the generated images. Guided by the LLM agent, we propose a tuning-free multi-concept customization model and a layout-to-image generation model as the tools for concept composition, and a local image editing method as the tool to interact with the agent for verification. The scene layout controls the image generation process among these tools to prevent confusion among multiple objects. Extensive experiments demonstrate the superiority of our approach for compositional text-to-image generation: CompAgent achieves more than 10\% improvement on T2I-CompBench, a comprehensive benchmark for open-world compositional T2I generation. The extension to various related tasks also illustrates the flexibility of our CompAgent for potential applications.

Yunzhong Hou, Zhongdao Wang, Shengjin Wang et al.

Data associations in multi-target multi-camera tracking (MTMCT) usually estimate affinity directly from re-identification (re-ID) feature distances. However, we argue that it might not be the best choice given the difference in matching scopes between re-ID and MTMCT problems. Re-ID systems focus on global matching, which retrieves targets from all cameras and all times. In contrast, data association in tracking is a local matching problem, since its candidates only come from neighboring locations and time frames. In this paper, we design experiments to verify such misfit between global re-ID feature distances and local matching in tracking, and propose a simple yet effective approach to adapt affinity estimations to corresponding matching scopes in MTMCT. Instead of trying to deal with all appearance changes, we tailor the affinity metric to specialize in ones that might emerge during data associations. To this end, we introduce a new data sampling scheme with temporal windows originally used for data associations in tracking. Minimizing the mismatch, the adaptive affinity module brings significant improvements over global re-ID distance, and produces competitive performance on CityFlow and DukeMTMC datasets.

Yuchi Liu, Zhongdao Wang, Tom Gedeon et al.

This paper does not contain technical novelty but introduces our key discoveries in a data generation protocol, a database and insights. We aim to address the lack of large-scale datasets in micro-expression (MiE) recognition due to the prohibitive cost of data collection, which renders large-scale training less feasible. To this end, we develop a protocol to automatically synthesize large scale MiE training data that allow us to train improved recognition models for real-world test data. Specifically, we discover three types of Action Units (AUs) that can constitute trainable MiEs. These AUs come from real-world MiEs, early frames of macro-expression videos, and the relationship between AUs and expression categories defined by human expert knowledge. With these AUs, our protocol then employs large numbers of face images of various identities and an off-the-shelf face generator for MiE synthesis, yielding the MiE-X dataset. MiE recognition models are trained or pre-trained on MiE-X and evaluated on real-world test sets, where very competitive accuracy is obtained. Experimental results not only validate the effectiveness of the discovered AUs and MiE-X dataset but also reveal some interesting properties of MiEs: they generalize across faces, are close to early-stage macro-expressions, and can be manually defined.

Zhongdao Wang, Hengshuang Zhao, Ya-Li Li et al.

Tracking objects of interest in a video is one of the most popular and widely applicable problems in computer vision. However, with the years, a Cambrian explosion of use cases and benchmarks has fragmented the problem in a multitude of different experimental setups. As a consequence, the literature has fragmented too, and now novel approaches proposed by the community are usually specialised to fit only one specific setup. To understand to what extent this specialisation is necessary, in this work we present UniTrack, a solution to address five different tasks within the same framework. UniTrack consists of a single and task-agnostic appearance model, which can be learned in a supervised or self-supervised fashion, and multiple ``heads'' that address individual tasks and do not require training. We show how most tracking tasks can be solved within this framework, and that the same appearance model can be successfully used to obtain results that are competitive against specialised methods for most of the tasks considered. The framework also allows us to analyse appearance models obtained with the most recent self-supervised methods, thus extending their evaluation and comparison to a larger variety of important problems.

Yuchi Liu, Zhongdao Wang, Xiangxin Zhou et al.

Association, aiming to link bounding boxes of the same identity in a video sequence, is a central component in multi-object tracking (MOT). To train association modules, e.g., parametric networks, real video data are usually used. However, annotating person tracks in consecutive video frames is expensive, and such real data, due to its inflexibility, offer us limited opportunities to evaluate the system performance w.r.t changing tracking scenarios. In this paper, we study whether 3D synthetic data can replace real-world videos for association training. Specifically, we introduce a large-scale synthetic data engine named MOTX, where the motion characteristics of cameras and objects are manually configured to be similar to those in real-world datasets. We show that compared with real data, association knowledge obtained from synthetic data can achieve very similar performance on real-world test sets without domain adaption techniques. Our intriguing observation is credited to two factors. First and foremost, 3D engines can well simulate motion factors such as camera movement, camera view and object movement, so that the simulated videos can provide association modules with effective motion features. Second, experimental results show that the appearance domain gap hardly harms the learning of association knowledge. In addition, the strong customization ability of MOTX allows us to quantitatively assess the impact of motion factors on MOT, which brings new insights to the community.

Zhongdao Wang, Jingwei Zhang, Liang Zheng et al.

This paper proposes a self-supervised learning method for the person re-identification (re-ID) problem, where existing unsupervised methods usually rely on pseudo labels, such as those from video tracklets or clustering. A potential drawback of using pseudo labels is that errors may accumulate and it is challenging to estimate the number of pseudo IDs. We introduce a different unsupervised method that allows us to learn pedestrian embeddings from raw videos, without resorting to pseudo labels. The goal is to construct a self-supervised pretext task that matches the person re-ID objective. Inspired by the \emph{data association} concept in multi-object tracking, we propose the \textbf{Cyc}le \textbf{As}sociation (\textbf{CycAs}) task: after performing data association between a pair of video frames forward and then backward, a pedestrian instance is supposed to be associated to itself. To fulfill this goal, the model must learn a meaningful representation that can well describe correspondences between instances in frame pairs. We adapt the discrete association process to a differentiable form, such that end-to-end training becomes feasible. Experiments are conducted in two aspects: We first compare our method with existing unsupervised re-ID methods on seven benchmarks and demonstrate CycAs' superiority. Then, to further validate the practical value of CycAs in real-world applications, we perform training on self-collected videos and report promising performance on standard test sets.

Yifan Sun, Changmao Cheng, Yuhan Zhang et al.

This paper provides a pair similarity optimization viewpoint on deep feature learning, aiming to maximize the within-class similarity $s_p$ and minimize the between-class similarity $s_n$. We find a majority of loss functions, including the triplet loss and the softmax plus cross-entropy loss, embed $s_n$ and $s_p$ into similarity pairs and seek to reduce $(s_n-s_p)$. Such an optimization manner is inflexible, because the penalty strength on every single similarity score is restricted to be equal. Our intuition is that if a similarity score deviates far from the optimum, it should be emphasized. To this end, we simply re-weight each similarity to highlight the less-optimized similarity scores. It results in a Circle loss, which is named due to its circular decision boundary. The Circle loss has a unified formula for two elemental deep feature learning approaches, i.e. learning with class-level labels and pair-wise labels. Analytically, we show that the Circle loss offers a more flexible optimization approach towards a more definite convergence target, compared with the loss functions optimizing $(s_n-s_p)$. Experimentally, we demonstrate the superiority of the Circle loss on a variety of deep feature learning tasks. On face recognition, person re-identification, as well as several fine-grained image retrieval datasets, the achieved performance is on par with the state of the art.

Yunzhong Hou, Liang Zheng, Zhongdao Wang et al.

Multi-target multi-camera tracking (MTMCT) systems track targets across cameras. Due to the continuity of target trajectories, tracking systems usually restrict their data association within a local neighborhood. In single camera tracking, local neighborhood refers to consecutive frames; in multi-camera tracking, it refers to neighboring cameras that the target may appear successively. For similarity estimation, tracking systems often adopt appearance features learned from the re-identification (re-ID) perspective. Different from tracking, re-ID usually does not have access to the trajectory cues that can limit the search space to a local neighborhood. Due to its global matching property, the re-ID perspective requires to learn global appearance features. We argue that the mismatch between the local matching procedure in tracking and the global nature of re-ID appearance features may compromise MTMCT performance. To fit the local matching procedure in MTMCT, in this work, we introduce locality aware appearance metric (LAAM). Specifically, we design an intra-camera metric for single camera tracking, and an inter-camera metric for multi-camera tracking. Both metrics are trained with data pairs sampled from their corresponding local neighborhoods, as opposed to global sampling in the re-ID perspective. We show that the locally learned metrics can be successfully applied on top of several globally learned re-ID features. With the proposed method, we report new state-of-the-art performance on the DukeMTMC dataset, and a substantial improvement on the CityFlow dataset.

Lanqing He, Zhongdao Wang, Yali Li et al.

The softmax loss and its variants are widely used as objectives for embedding learning, especially in applications like face recognition. However, the intra- and inter-class objectives in the softmax loss are entangled, therefore a well-optimized inter-class objective leads to relaxation on the intra-class objective, and vice versa. In this paper, we propose to dissect the softmax loss into independent intra- and inter-class objective (D-Softmax). With D-Softmax as objective, we can have a clear understanding of both the intra- and inter-class objective, therefore it is straightforward to tune each part to the best state. Furthermore, we find the computation of the inter-class objective is redundant and propose two sampling-based variants of D-Softmax to reduce the computation cost. Training with regular-scale data, experiments in face verification show D-Softmax is favorably comparable to existing losses such as SphereFace and ArcFace. Training with massive-scale data, experiments show the fast variants of D-Softmax significantly accelerates the training process (such as 64x) with only a minor sacrifice in performance, outperforming existing acceleration methods of softmax in terms of both performance and efficiency.

Zhongdao Wang, Liang Zheng, Yali Li et al.

In this paper, we present an accurate and scalable approach to the face clustering task. We aim at grouping a set of faces by their potential identities. We formulate this task as a link prediction problem: a link exists between two faces if they are of the same identity. The key idea is that we find the local context in the feature space around an instance (face) contains rich information about the linkage relationship between this instance and its neighbors. By constructing sub-graphs around each instance as input data, which depict the local context, we utilize the graph convolution network (GCN) to perform reasoning and infer the likelihood of linkage between pairs in the sub-graphs. Experiments show that our method is more robust to the complex distribution of faces than conventional methods, yielding favorably comparable results to state-of-the-art methods on standard face clustering benchmarks, and is scalable to large datasets. Furthermore, we show that the proposed method does not need the number of clusters as prior, is aware of noises and outliers, and can be extended to a multi-view version for more accurate clustering accuracy.

Zhongdao Wang, Liang Zheng, Shengjin Wang

Feature fusion is a commonly used strategy in image retrieval tasks, which aggregates the matching responses of multiple visual features. Feasible sets of features can be either descriptors (SIFT, HSV) for an entire image or the same descriptor for different local parts (face, body). Ideally, the to-be-fused heterogeneous features are pre-assumed to be discriminative and complementary to each other. However, the effectiveness of different features varies dramatically according to different queries. That is to say, for some queries, a feature may be neither discriminative nor complementary to existing ones, while for other queries, the feature suffices. As a result, it is important to estimate the effectiveness of features in a query-adaptive manner. To this end, this article proposes a new late fusion scheme at the score level. We base our method on the observation that the sorted score curves contain patterns that describe their effectiveness. For example, an "L"-shaped curve indicates that the feature is discriminative while a gradually descending curve suggests a bad feature. As such, this paper introduces a query-adaptive late fusion pipeline. In the hand-crafted version, it can be an unsupervised approach to tasks like particular object retrieval. In the learning version, it can also be applied to supervised tasks like person recognition and pedestrian retrieval, based on a trainable neural module. Extensive experiments are conducted on two object retrieval datasets and one person recognition dataset. We show that our method is able to highlight the good features and suppress the bad ones, is resilient to distractor features, and achieves very competitive retrieval accuracy compared with the state of the art. In an additional person re-identification dataset, the application scope and limitation of the proposed method are studied.