Recent advancements in image generation models enable the creation of high-quality images and targeted modifications based on textual instructions. Some models even support multimodal complex guidance and demonstrate robust task generalization capabilities. However, they still fall short of meeting the nuanced, professional demands of designers. To bridge this gap, we introduce IDEA-Bench, a comprehensive benchmark designed to advance image generation models toward applications with robust task generalization. IDEA-Bench comprises 100 professional image generation tasks and 275 specific cases, categorized into five major types based on the current capabilities of existing models. Furthermore, we provide a representative subset of 18 tasks with enhanced evaluation criteria to facilitate more nuanced and reliable evaluations using Multimodal Large Language Models (MLLMs). By assessing models’ ability to comprehend and execute novel, complex tasks, IDEA-Bench paves the way toward the development of generative models with autonomous and versatile visual generation capabilities.

This work presents AnyDoor, a diffusion-based image generator with the power to teleport target objects to new scenes at user-specified locations with desired shapes. Instead of tuning parameters for each object, our model is trained only once and effortlessly generalizes to diverse object-scene combinations at the inference stage. Such a challenging zero-shot setting requires an adequate characterization of a certain object. To this end, we leverage the powerful self-supervised image encoder (i.e., DINOv2) to extract the discriminative dentity feature of the target object. Besides, we complement the identity feature with detail features, which are carefully designed to maintain appearance details yet allow versatile local variations (e.g., lighting, orientation, posture, etc.), supporting the object in favorably blending with different surroundings. We further propose to borrow knowledge from video datasets, where we can observe various forms (i.e., along the time axis) of a single object, leading to stronger model generalizability and robustness. Starting from the task of object insertion, we further extend the framework of AnyDoor to a general solution with region-to-region image reference. With the different definitions of the source region and target region, the tasks of object insertion, object removal, and image variation could be integrated into one model without introducing extra parameters. In addition, we investigate incorporating other conditions like the mask, pose skeleton, and depth map as additional guidance to achieve more controllable generation.

This work presents AnyDoor, a diffusion-based image generator with the power to teleport target objects to new scenes at user-specified locations with desired shapes. Instead of tuning parameters for each object, our model is trained only once and effortlessly generalizes to diverse object-scene combinations at the inference stage. Such a challenging zero-shot setting requires an adequate characterization of a certain object. To this end, we complement the commonly used identity feature with detail features, which are carefully designed to maintain appearance details yet allow versatile local variations (e.g., lighting, orientation, posture, etc.), supporting the object in favorably blending with different surroundings. We further propose to borrow knowledge from video datasets, where we can observe various forms (i.e., along the time axis) of a single object, leading to stronger model generalizability and robustness. Extensive experiments demonstrate the superiority of our approach over existing alternatives as well as its great potential in real-world applications, such as virtual try-on, shape editing, and object swapping. Code is released at github.com/ali-vilab/AnyDoor.

Diffusion models, which employ stochastic differential equations to sample images through integrals, have emerged as a dominant class of generative models. However, the rationality of the diffusion process itself receives limited attention, leaving the question of whether the problem is well-posed and well-conditioned. In this paper, we uncover a vexing propensity of diffusion models: they frequently exhibit the infinite Lipschitz near the zero point of timesteps. We provide theoretical proofs to illustrate the presence of infinite Lipschitz constants and empirical results to confirm it. The Lipschitz singularities pose a threat to the stability and accuracy during both the training and inference processes of diffusion models. Therefore, the mitigation of Lipschitz singularities holds great potential for enhancing the performance of diffusion models. To address this challenge, we propose a novel approach, dubbed E-TSDM, which alleviates the Lipschitz singularities of the diffusion model near the zero point. Remarkably, our technique yields a substantial improvement in performance. Moreover, as a byproduct of our method, we achieve a dramatic reduction in the Fréchet Inception Distance of acceleration methods relying on network Lipschitz, including DDIM and DPM-Solver, by over 33\%. Extensive experiments on diverse datasets validate our theory and method. Our work may advance the understanding of the general diffusion process, and also provide insights for the design of diffusion models.

Diffusion models, which learn to reverse a signal destruction process to generate new data, typically require the signal at each step to have the same dimension. We argue that, considering the spatial redundancy in image signals, there is no need to maintain a high dimensionality in the evolution process, especially in the early generation phase. To this end, we make a theoretical generalization of the forward diffusion process via signal decomposition. Concretely, we manage to decompose an image into multiple orthogonal components and control the attenuation of each component when perturbing the image. That way, along with the noise strength increasing, we are able to diminish those inconsequential components and thus use a lower-dimensional signal to represent the source, barely losing information. Such a reformulation allows to vary dimensions in both training and inference of diffusion models. Extensive experiments on a range of datasets suggest that our approach substantially reduces the computational cost and achieves on-par or even better synthesis performance compared to baseline methods. We also show that our strategy facilitates high-resolution image synthesis and improves FID of diffusion model trained on FFHQ at$1024\times 1024$resolution from 52.40 to 10.46. Code is available at https://github.com/damo-vilab/dvdp.

Vector-quantized image modeling has shown great potential in synthesizing high-quality images. However, generating high-resolution images remains a challenging task due to the quadratic computational overhead of the self-attention process. In this study, we seek to explore a more efficient two-stage framework for high-resolution image generation with improvements in the following three aspects. (1) Based on the observation that the first quantization stage has solid local property, we employ a local attention-based quantization model instead of the global attention mechanism used in previous methods, leading to better efficiency and reconstruction quality. (2) We emphasize the importance of multi-grained feature interaction during image generation and introduce an efficient attention mechanism that combines global attention (long-range semantic consistency within the whole image) and local attention (fined-grained details). This approach results in faster generation speed, higher generation fidelity, and improved resolution. (3) We propose a new generation pipeline incorporating autoencoding training and autoregressive generation strategy, demonstrating a better paradigm for image synthesis. Extensive experiments demonstrate the superiority of our approach in high-quality and high-resolution image reconstruction and generation.

Recent large-scale generative models learned on big data are capable of synthesizing incredible images yet suffer from limited controllability. This work offers a new generation paradigm that allows flexible control of the output image, such as spatial layout and palette, while maintaining the synthesis quality and model creativity. With compositionality as the core idea, we first decompose an image into representative factors, and then train a diffusion model with all these factors as the conditions to recompose the input. At the inference stage, the rich intermediate representations work as composable elements, leading to a huge design space (i.e., exponentially proportional to the number of decomposed factors) for customizable content creation. It is noteworthy that our approach, which we call Composer, supports various levels of conditions, such as text description as the global information, depth map and sketch as the local guidance, color histogram for low-level details, etc. Besides improving controllability, we confirm that Composer serves as a general framework and facilitates a wide range of classical generative tasks without retraining. Code and models will be made available.

Target tracking, the essential ability of the human visual system, has been simulated by computer vision tasks. However, existing trackers perform well in austere experimental environments but fail in challenges like occlusion and fast motion. The massive gap indicates that researches only measure tracking performance rather than intelligence. How to scientifically judge the intelligence level of trackers? Distinct from decision-making problems, lacking three requirements (a challenging task, a fair environment, and a scientific evaluation procedure) makes it strenuous to answer the question. In this article, we first propose the global instance tracking (GIT) task, which is supposed to search an arbitrary user-specified instance in a video without any assumptions about camera or motion consistency, to model the human visual tracking ability. Whereafter, we construct a high-quality and large-scale benchmark VideoCube to create a challenging environment. Finally, we design a scientific evaluation procedure using human capabilities as the baseline to judge tracking intelligence. Additionally, we provide an online platform with toolkit and an updated leaderboard. Although the experimental results indicate a definite gap between trackers and humans, we expect to take a step forward to generate authentic human-like trackers. The database, toolkit, evaluation server, and baseline results are available at http://videocube.aitestunion.com.

Fashion is the form we express ourselves to the world and has grown into one of the largest industries in the world. Despite the significant evolvement of the fashion industry over the past decades, it is still a great challenge to respond to the diverse preferences of a large number of different consumers in time and accurately. To deal with the problem, we present an innovative demonstration of a trend-driven fashion design system using deep generative modeling, which enables automatic fashion design and editing based on trend reports. Our system consists of three components, including trend-driven fashion design, interactive fashion editing, and popularity estimation. The system offers a unified framework for the mass production of fashion designs that conform to the trend, which helps businesses better respond to market demands.

Recent works show the possibility of transferring the CLIP (Contrastive Language-Image Pretraining) model for video-text retrieval with promising performance. However, due to the domain gap between static images and videos, CLIP-based video-text retrieval models with interaction-based matching perform far worse than models with representation-based matching. In this paper, we propose a novel image animation strategy to transfer the image-text CLIP model to video-text retrieval effectively. By imitating the video shooting components, we convert widely used image-language corpus to synthesized video-text data for pretraining. To reduce the time complexity of interaction matching, we further propose a coarse to fine framework which consists of dual encoders for fast candidates searching and a cross-modality interaction module for fine-grained re-ranking. The coarse to fine framework with the synthesized video-text pretraining provides significant gains in retrieval accuracy while preserving efficiency. Comprehensive experiments conducted on MSR-VTT, MSVD, and VATEX datasets demonstrate the effectiveness of our approach.

In its most basic form, advertising video production communicates a message about a product or service to the public. In the age of digital marketing, where the most popular way to connect with audiences is through advertising videos. However, advertising video production is a costly and complicated process from creation, material shooting, editing to the final commercial video. Therefore, producing qualified advertising videos is a capital and talent-intensive task, which poses a huge challenge for start-ups or inexperienced ad creators. paper proposes an intelligent advertising video production system driven by multi-modal retrieval, which only requires the input of descriptive copy. This system can automatically generate scripts, then extract key queries, retrieve related short video materials in the video library, and finally synthesize short advertising videos. The whole process minimizes human input, greatly reduces the threshold for advertising video production and greatly improves output and efficiency. It has a modular design to encourage the study of new multi-modal algorithms, which can be evaluated in batch mode. It can also integrate with a user interface, which allows user studies and data collection in an interactive mode, where the back end can be fully algorithmic or a wizard of oz setup. The proposed system has been fully verified and has broad prospects in the production of short videos for commodity advertisements within Alibaba.

This paper presents a simple unsupervised visual representation learning method with a pretext task of discriminating all images in a dataset using a parametric, instance-level classifier. The overall framework is a replica of a supervised classification model, where semantic classes (e.g., dog, bird, and ship) are replaced by instance IDs. However, scaling up the classification task from thousands of semantic labels to millions of instance labels brings specific challenges including 1) the large-scale softmax computation; 2) the slow convergence due to the infrequent visiting of instance samples; and 3) the massive number of negative classes that can be noisy. This work presents several novel techniques to handle these difficulties. First, we introduce a hybrid parallel training framework to make large-scale training feasible. Second, we present a raw-feature initialization mechanism for classification weights, which we assume offers a contrastive prior for instance discrimination and can clearly speed up converge in our experiments. Finally, we propose to smooth the labels of a few hardest classes to avoid optimizing over very similar negative pairs. While being conceptually simple, our framework achieves competitive or superior performance compared to state-of-the-art unsupervised approaches, i.e., SimCLR, MoCoV2, and PIC under ImageNet linear evaluation protocol and on several downstream visual tasks, verifying that full instance classification is a strong pretraining technique for many semantic visual tasks.

A key challenge in self-supervised video representation learning is how to effectively capture motion information besides context bias. While most existing works implicitly achieve this with video-specific pretext tasks (e.g., predicting clip orders, time arrows, and paces), we develop a method that explicitly decouples motion supervision from context bias through a carefully designed pretext task. Specifically, we take the key frames and motion vectors in compressed videos (e.g., in H.264 format) as the supervision sources for context and motion, respectively, which can be efficiently extracted at over 500 fps on CPU. Then we design two pretext tasks that are jointly optimized: a context matching task where a pairwise contrastive loss is cast between video clip and key frame features; and a motion prediction task where clip features, passed through an encoderdecoder network, are used to estimate motion features in a near future. These two tasks use a shared video backbone and separate MLP heads. Experiments show that our approach improves the quality of the learned video representation over previous works, where we obtain absolute gains of 16.0% and 11.1% in video retrieval recall on UCF101 and HMDB51, respectively. Moreover, we find the motion prediction to be a strong regularization for video networks, where using it as an auxiliary task improves the accuracy of action recognition with a margin of 7.4% ∼ 13.8%.

Videos grow to be one of the largest mediums on the Internet. E-commerce platforms like Alibaba need to process millions of video data across multimedia (e.g., visual, audio, image, and text) and on a variety of tasks (e.g., retrieval, tagging, and summary) every day. In this work, we aim to develop a once and for all pretraining technique for diverse modalities and downstream tasks. To achieve this, we make the following contributions: (1) We propose a self-supervised multi-modal co-training framework. It takes cross-modal pseudo-label consistency as the supervision and can jointly learn representations of multiple modalities. (2) We introduce several novel techniques (e.g., sliding-window subset sampling, coarse-to-fine clustering, fast spatial-temporal convolution and parallel data transmission and processing) to optimize the training process, making billion-scale stable training feasible. (3) We construct a large-scale multi-modal dataset consisting of 1.4 billion videos (~0.5 PB) and train our framework on it. The training takes only 4.6 days on an in-house 256 GPUs cluster, and it simultaneously produces pretrained video, audio, image, motion, and text networks. (4) Finetuning from our pretrained models, we obtain significant performance gains and faster convergence on diverse multimedia tasks at Alibaba. Furthermore, we also validate the learned representation on public datasets. Despite the domain gap between our commodity-centric pretraining and the action-centric evaluation data, we show superior results against state-of-the-arts.

We introduce here a large tracking database that offers an unprecedentedly wide coverage of common moving objects in the wild, called GOT-10k. Specifically, GOT-10k is built upon the backbone of WordNet structure [1] and it populates the majority of over 560 classes of moving objects and 87 motion patterns, magnitudes wider than the most recent similar-scale counterparts [19], [20], [23], [26]. By releasing the large high-diversity database, we aim to provide a unified training and evaluation platform for the development of class-agnostic, generic purposed short-term trackers. The features of GOT-10k and the contributions of this article are summarized in the following. (1) GOT-10k offers over 10,000 video segments with more than 1.5 million manually labeled bounding boxes, enabling unified training and stable evaluation of deep trackers. (2) GOT-10k is by far the first video trajectory dataset that uses the semantic hierarchy of WordNet to guide class population, which ensures a comprehensive and relatively unbiased coverage of diverse moving objects. (3) For the first time, GOT-10k introduces the one-shot protocol for tracker evaluation, where the training and test classes are zero-overlapped. The protocol avoids biased evaluation results towards familiar objects and it promotes generalization in tracker development. (4) GOT-10k offers additional labels such as motion classes and object visible ratios, facilitating the development of motion-aware and occlusion-aware trackers. (5) We conduct extensive tracking experiments with 39 typical tracking algorithms and their variants on GOT-10k and analyze their results in this paper. (6) Finally, we develop a comprehensive platform for the tracking community that offers full-featured evaluation toolkits, an online evaluation server, and a responsive leaderboard. The annotations of GOT-10k's test data are kept private to avoid tuning parameters on it.

Studies have shown that contextual information can promote the robustness of trackers. However, trackers based on convolutional neural networks (CNNs) only capture local features, which limits their performance. We propose a novel relevant context block (RCB), which employs graph convolutional networks to capture the relevant context. In particular, it selects the$k$largest contributors as nodes for each query position (unit) that contain meaningful and discriminative contextual information and updates the nodes by aggregating the differences between the query position and its contributors. This operation can be easily incorporated into the existing networks and can be easily end-to-end trained using a standard backpropagation algorithm. To verify the effectiveness of RCB, we apply it to two trackers, SiamFC and GlobalTrack, respectively, and the two improved trackers are referred to as Siam-RCB and GlobalTrack-RCB. Extensive experiments on OTB, VOT, UAV123, LaSOT, TrackingNet, OxUvA, and VOT2018LT show the superiority of our method. For example, our Siam-RCB outperforms SiamFC by a very large margin (up to 11.2% in the success score and 7.8% in the precision score) on the OTB-100 benchmark.

A key capability of a long-term tracker is to search for targets in very large areas (typically the entire image) to handle possible target absences or tracking failures. However, currently there is a lack of such a strong baseline for global instance search. In this work, we aim to bridge this gap. Specifically, we propose GlobalTrack, a pure global instance search based tracker that makes no assumption on the temporal consistency of the target's positions and scales. GlobalTrack is developed based on two-stage object detectors, and it is able to perform full-image and multi-scale search of arbitrary instances with only a single query as the guide. We further propose a cross-query loss to improve the robustness of our approach against distractors. With no online learning, no punishment on position or scale changes, no scale smoothing and no trajectory refinement, our pure global instance search based tracker achieves comparable, sometimes much better performance on four large-scale tracking benchmarks (i.e., 52.1% AUC on LaSOT, 63.8% success rate on TLP, 60.3% MaxGM on OxUvA and 75.4% normalized precision on TrackingNet), compared to state-of-the-art approaches that typically require complex post-processing. More importantly, our tracker runs without cumulative errors, i.e., any type of temporary tracking failures will not affect its performance on future frames, making it ideal for long-term tracking. We hope this work will be a strong baseline for long-term tracking and will stimulate future works in this area.

Object detection models have been a source of inspiration for many tracking-by-detection algorithms over the past decade. Recent deep trackers borrow designs or modules from the latest object detection methods, such as bounding box regression, RPN and ROI pooling, and can deliver impressive performance. In this paper, instead of redesigning a new tracking-by-detection algorithm, we aim to explore a general framework for building trackers directly upon almost any advanced object detector. To achieve this, three key gaps must be bridged: (1) Object detectors are class-specific, while trackers are class-agnostic. (2) Object detectors do not differentiate intra-class instances, while this is a critical capability of a tracker. (3) Temporal cues are important for stable long-term tracking while they are not considered in still-image detectors. To address the above issues, we first present a simple target-guidance module for guiding the detector to locate target-relevant objects. Then a meta-learner is adopted for the detector to fast learn and adapt a target-distractor classifier online. We further introduce an anchored updating strategy to alleviate the problem of overfitting. The framework is instantiated on SSD and FasterRCNN, the typical one- and two-stage detectors, respectively. Experiments on OTB, UAV123 and NfS have verified our framework and show that our trackers can benefit from deeper backbone networks, as opposed to many recent trackers.

In this paper, we present a novel part-based visual tracking method from the perspective of probability sampling. Specifically, we represent the target by a part space with two online learned probabilities to capture the structure of the target. The proposal distribution memorizes the historical performance of different parts, and it is used for the first round of part selection. The acceptance probability validates the specific tracking stability of each part in a frame, and it determines whether to accept its vote or to reject it. By doing this, we transform the complex online part selection problem into a probability learning one, which is easier to tackle. The observation model of each part is constructed by an improved supervised descent method and is learned in an incremental manner. Experimental results on two benchmarks demonstrate the competitive performance of our tracker against state-of-the-art methods.

How to effectively organize local descriptors to build a global representation has a critical impact on the performance of vision tasks. Recently, local sparse representation has been successfully applied to visual tracking, owing to its discriminative nature and robustness against local noise and partial occlusions. Local sparse codes computed with a template actually form a three-order tensor according to their original layout, although most existing pooling operators convert the codes to a vector by concatenating or computing statistics on them. We argue that, compared to pooling vectors, the tensor form could deliver more intrinsic structural information for the target appearance, and can also avoid high dimensionality learning problems suffered in concatenation-based pooling methods. Therefore, in this paper, we propose to represent target templates and candidates directly with sparse coding tensors, and build the appearance model by incrementally learning on these tensors. We propose a discriminative framework to further improve robustness of our method against drifting and environmental noise. Experiments on a recent comprehensive benchmark indicate that our method performs better than state-of-the-art trackers.

Most existing tracking algorithms do not explicitly consider the motion blur contained in video sequences, which degrades their performance in real-world applications where motion blur often occurs. In this paper, we propose to solve the motion blur problem in visual tracking in a unified framework. Specifically, a joint blur state estimation and multi-task reverse sparse learning framework are presented, where the closed-form solution of blur kernel and sparse code matrix is obtained simultaneously. The reverse process considers the blurry candidates as dictionary elements, and sparsely represents blurred templates with the candidates. By utilizing the information contained in the sparse code matrix, an efficient likelihood model is further developed, which quickly excludes irrelevant candidates and narrows the particle scale down. Experimental results on the challenging benchmarks show that our method performs well against the state-of-the-art trackers.

Recently, local sparse representation (LSR) has been successfully applied in visual tracking, owing to its discriminative nature and robustness against local noise and occlusions. It is note worthy that local sparse codes computed with a template form a 3-order tensor of their original layout, although most pooling operators convert it to a vector by concatenating or computing statistics on it. As compared to pooling vectors, tensor form could deliver more informative and structured representation for target appearance, and can also avoid high dimensionality learning problem suffered in concatenating pooling based methods. Motivated by above ideas, in this paper, we propose to represent target templates directly with sparse coding tensors, and build the appearance model by incrementally learning on these tensors. We further propose a discriminative framework to improve robustness against drifting and environment noise. Experiments on a recent comprehensive benchmark indicate that our method outperforms state-of-the-art trackers.