DiT models have achieved great success in text-to-video generation, leveraging their scalability in model capacity and data scale. High content and motion fidelity aligned with text prompts, however, often require large model parameters and a substantial number of function evaluations (NFEs). Realistic and visually appealing details are typically reflected in high-resolution outputs, further amplifying computational demands—especially for single-stage DiT models. To address these challenges, we propose a novel two-stage framework, FlashVideo, which strategically allocates model capacity and NFEs across stages to balance generation fidelity and quality. In the first stage, prompt fidelity is prioritized through a low-resolution generation process utilizing large parameters and sufficient NFEs to enhance computational efficiency. The second stage achieves a nearly straight ODE trajectory between low and high resolutions via flow matching, effectively generating fine details and fixing artifacts with minimal NFEs. To ensure a seamless connection between the two independently trained stages during inference, we carefully design degradation strategies during the second-stage training. Quantitative and visual results demonstrate that FlashVideo achieves state-of-the-art high-resolution video generation with superior computational efficiency. Additionally, the two-stage design enables users to preview the initial output and accordingly adjust the prompt before committing to full-resolution generation, thereby significantly reducing computational costs and wait times as well as enhancing commercial viability.

This paper introduces Goku, a state-of-the-art family of joint image-and-video generation models leveraging rectified flow Transformers to achieve industry-leading performance. We detail the foundational elements enabling high-quality visual generation, including the data curation pipeline, model architecture design, flow formulation, and advanced infrastructure for efficient and robust large-scale training. The Goku models demonstrate superior performance in both qualitative and quantitative evaluations, setting new benchmarks across major tasks. Specifically, Goku achieves 0.76 on GenEval and 83.65 on DPG-Bench for text-to-image generation, and 84.85 on VBench for text-to-video tasks. We believe that this work provides valuable insights and practical advancements for the research community in developing joint image-and-video generation models.

We present Infinity, a Bitwise Visual AutoRegressive Modeling capable of generating high-resolution, photorealistic images following language instruction. Infinity refactors visual autoregressive model under a bitwise token prediction framework with an infinite-vocabulary classifier and bit-wise self-correction mechanism. By theoretically expanding the tokenizer vocabulary size to infinity in Transformer, our method significantly unleashes powerful scaling capabilities to infinity compared to vanilla VAR. Extensive experiments indicate Infinity outperforms AutoRegressive Text-to-Image models by large margins, matches or surpasses leading diffusion models. Without extra optimization, Infinity generates a 1024×1024 image in 0.8s, 2.6× faster than SD3-Medium, making it the fastest Text-to-Image model. All the code and models are available to promote further exploration of Infinity for visual generation.

We present TokenFlow, a novel unified image tokenizer that bridges the long-standing gap between multimodal understanding and generation. Prior research attempt to employ a single reconstruction-targeted Vector Quantization (VQ) encoder for unifying these two tasks. We observe that understanding and generation require fundamentally different granularities of visual information. This leads to a critical trade-off, particularly compromising performance in multimodal understanding tasks. TokenFlow addresses this challenge through an innovative dual-codebook architecture that decouples semantic and pixel-level feature learning while maintaining their alignment via a shared mapping mechanism. This design enables direct access to both high-level semantic representations crucial for understanding tasks and fine-grained visual features essential for generation through shared indices. Our extensive experiments demonstrate TokenFlow’s superiority across multiple dimensions. Leveraging TokenFlow, we demonstrate for the first time that discrete visual input can surpass LLaVA-1.5 13B in understanding performance, achieving a 7.2% average improvement. For image reconstruction, we achieve a strong FID score of 0.63 at 384×384 resolution. Moreover, TokenFlow establishes state-of-the-art performance in autoregressive image generation with a GenEval score of 0.55 at 256×256 resolution, achieving comparable results to SDXL. Our code and models are released at https://github.com/ByteFlow-Ai/TokenFlow.

We introduce InfinityStar, a unified spacetime autoregressive framework for high-resolution image and dynamic video synthesis. Building on the recent success of autoregressive modeling in both vision and language, our purely discrete approach jointly captures spatial and temporal dependencies within a single architecture. This unified design naturally supports a variety of generation tasks such as text-to-image, text-to-video, image-to-video, and long-duration video synthesis via straightforward temporal autoregression. Through extensive experiments, InfinityStar scores 83.74 on VBench, outperforming all autoregressive models by large margins, even surpassing diffusion competitors like HunyuanVideo. Without extra optimizations, our model generates a 5s, 720p video approximately 10$\times$ faster than leading diffusion-based methods. To our knowledge, InfinityStar is the first discrete autoregressive video generator capable of producing industrial-level 720p videos. We release all code and models to foster further research in efficient, high-quality video generation.

Visual generative and understanding models typically rely on distinct tokenizers to process images, presenting a key challenge for unifying them within a single framework. Recent studies attempt to address this by connecting the training of VQVAE (for autoregressive generation) and CLIP (for understanding) to build a unified tokenizer. However, directly combining these training objectives has been observed to cause severe loss conflicts. In this paper, we show that reconstruction and semantic supervision do not inherently conflict. Instead, the underlying bottleneck stems from limited representational capacity of discrete token space. Building on these insights, we introduce UniTok, a unified tokenizer featuring a novel multi-codebook quantization mechanism that effectively scales up the vocabulary size and bottleneck dimension. In terms of final performance, UniTok sets a new record of 0.38 rFID and 78.6\% zero-shot accuracy on ImageNet. Besides, UniTok can be seamlessly integrated into MLLMs to unlock native visual generation capability, without compromising the understanding performance. Additionally, we show that UniTok favors cfg-free generation, reducing gFID from 14.6 to 2.5 on ImageNet 256$\times$256 benchmark. All codes and models have been made publicly available.

Building scalable vision-language models to learn from diverse, multimodal data remains an open challenge. In this paper, we introduce an Efficient Vision-languagE foundation model, namely EVE, which is one unified multimodal Transformer pre-trained solely by one unified pre-training task. Specifically, EVE encodes both vision and language within a shared Transformer network integrated with modality-aware sparse Mixture-of-Experts (MoE) modules, which capture modality-specific information by selectively switching to different experts. To unify pre-training tasks of vision and language, EVE performs masked signal modeling on image-text pairs to reconstruct masked signals, i.e., image pixels and text tokens, given visible signals. This simple yet effective pre-training objective accelerates training by 4x compared to the model pre-trained with Image-Text Contrastive and Image-Text Matching losses. Owing to the combination of the unified architecture and pre-training task, EVE is easy to scale up, enabling better downstream performance with fewer resources and faster training speed. Despite its simplicity, EVE achieves state-of-the-art performance on various vision-language downstream tasks, including visual question answering, visual reasoning, and image-text retrieval.

In recent research, significant attention has been devoted to the open-vocabulary object detection task, aiming to generalize beyond the limited number of classes labeled during training and detect objects described by arbitrary category names at inference. Compared with conventional object detection, open vocabulary object detection largely extends the object detection categories. However, it relies on calculating the similarity between image regions and a set of arbitrary category names with a pretrained vision-and-language model. This implies that, despite its open-set nature, the task still needs the predefined object categories during the inference stage. This raises the question: What if we do not have exact knowledge of object categories during inference? In this paper, we call such a new setting as generative open-ended object detection, which is a more general and practical problem. To address it, we formulate object detection as a generative problem and propose a simple framework named GenerateU, which can detect dense objects and generate their names in a free-form way. Particularly, we employ Deformable DETR as a region proposal generator with a language model translating visual regions to object names. To assess the free-form object detection task, we introduce an evaluation method designed to quantitatively measure the performance of generative out-comes. Extensive experiments demonstrate strong zero-shot detection performance of our GenerateU. For example, on the LVIS dataset, our GenerateU achieves comparable results to the open-vocabulary object detection method GLIP, even though the category names are not seen by GenerateU during inference. Code is available at: https://github.com/FoundationVision/GenerateU.

We present GLEE in this work, an object-level foundation model for locating and identifying objects in images and videos. Through a unified framework, GLEE accomplishes detection, segmentation, tracking, grounding, and identification of arbitrary objects in the open world scenario for various object perception tasks. Adopting a cohesive learning strategy, GLEE acquires knowledge from diverse data sources with varying supervision levels to formu-late general object representations, excelling in zero-shot transfer to new data and tasks. Specifically, we employ an image encoder, text encoder, and visual prompter to handle multimodal inputs, enabling to simultaneously solve various object-centric downstream tasks while maintaining state-of-the-art performance. Demonstrated through extensive training on over five million images from diverse benchmarks, GLEE exhibits remarkable versatility and improved generalization performance, efficiently tack-ling downstream tasks without the need for task-specific adaptation. By integrating large volumes of automatically labeled data, we further enhance its zero-shot generalization capabilities. Additionally, GLEE is capable of being integrated into Large Language Models, serving as a foun-dational model to provide universal object-level information for multimodal tasks. We hope that the versatility and universality of our method will mark a significant step in the development of efficient visual foundation models for AGI systems. The models and code are released at https://github.com/FoundationVision/GLEE.

Existing image-level self-supervised learning (SSL) methods pre-trained on natural scene data can have difficulty in adating to dense prediction tasks. However, scene images contain multiple varied instances. We devise two techniques to craft high-quality scene and instance views for instance-level SSL. Firstly, we leverage the prior from the image-level pre-trained model to discover the salient instances and the correspondence between cross-image instance pairs. Then, we augment multiple instances to compose the scene views, where both scene similarity and variance are promoted. Meanwhile, we encourage the scene and inner instances to align in feature space to model the scene-instance correlation. Experiments show its SOTA performance in dense prediction tasks.

We present Visual AutoRegressive modeling (VAR), a new generation paradigm that redefines the autoregressive learning on images as coarse-to-fine "next-scale prediction" or "next-resolution prediction", diverging from the standard raster-scan "next-token prediction". This simple, intuitive methodology allows autoregressive (AR) transformers to learn visual distributions fast and generalize well: VAR, for the first time, makes GPT-style AR models surpass diffusion transformers in image generation. On ImageNet 256x256 benchmark, VAR significantly improve AR baseline by improving Frechet inception distance (FID) from 18.65 to 1.73, inception score (IS) from 80.4 to 350.2, with around 20x faster inference speed. It is also empirically verified that VAR outperforms the Diffusion Transformer (DiT) in multiple dimensions including image quality, inference speed, data efficiency, and scalability. Scaling up VAR models exhibits clear power-law scaling laws similar to those observed in LLMs, with linear correlation coefficients near -0.998 as solid evidence. VAR further showcases zero-shot generalization ability in downstream tasks including image in-painting, out-painting, and editing. These results suggest VAR has initially emulated the two important properties of LLMs: Scaling Laws and zero-shot task generalization. We have released all models and codes to promote the exploration of AR/VAR models for visual generation and unified learning.

Tokenizer, serving as a translator to map the intricate visual data into a compact latent space, lies at the core of visual generative models. Based on the finding that existing tokenizers are tailored to either image or video inputs, this paper presents OmniTokenizer, a transformer-based tokenizer for joint image and video tokenization. OmniTokenizer is designed with a spatial-temporal decoupled architecture, which integrates window attention and causal attention for spatial and temporal modeling, respectively. To exploit the complementary nature of image and video data, we further propose a progressive training strategy, where OmniTokenizer is first trained on image data on a fixed resolution to develop the spatial encoding capacity and then jointly trained on image and video data on multiple resolutions to learn the temporal dynamics. OmniTokenizer, for the first time, handles both image and video inputs within a unified framework and proves the possibility of realizing their synergy. Extensive experiments demonstrate that OmniTokenizer achieves state-of-the-art (SOTA) reconstruction performance on various image and video datasets, e.g., 1.11 reconstruction FID on ImageNet and 42 reconstruction FVD on UCF-101, beating the previous SOTA methods by 13% and 26%, respectively. Additionally, we also show that when integrated with OmniTokenizer, both language model-based approaches and diffusion models can realize advanced visual synthesis performance, underscoring the superiority and versatility of our method.

Understanding the semantics of individual regions or patches of unconstrained images, such as open-world object detection, remains a critical yet challenging task in computer vision. Building on the success of powerful image-level vision-language (ViL) foundation models like CLIP, recent efforts have sought to harness their capabilities by either training a contrastive model from scratch with an extensive collection of region-label pairs or aligning the outputs of a detection model with image-level representations of region proposals. Despite notable progress, these approaches are plagued by computationally intensive training requirements, susceptibility to data noise, and deficiency in contextual information. To address these limitations, we explore the synergistic potential of off-the-shelf foundation models, leveraging their respective strengths in localization and semantics. We introduce a novel, generic, and efficient architecture, named RegionSpot, designed to integrate position-aware localization knowledge from a localization foundation model (e.g., SAM) with semantic information from a ViL model (e.g., CLIP). To fully exploit pretrained knowledge while minimizing training overhead, we keep both foundation models frozen, focusing optimization efforts solely on a lightweight attention-based knowledge integration module. Extensive experiments in open-world object recognition show that our RegionSpot achieves significant performance gain over prior alternatives, along with substantial computational savings (e.g., training our model with 3 million data in a single day using 8 V100 GPUs). RegionSpot outperforms GLIP-L by 2.9 in mAP on LVIS val set, with an even larger margin of 13.1 AP for more challenging and rare categories, and a 2.5 AP increase on ODinW. Furthermore, it exceeds GroundingDINO-L by 11.0 AP for rare categories on the LVIS minival set.

All instance perception tasks aim at finding certain objects specified by some queries such as category names, language expressions, and target annotations, but this complete field has been split into multiple independent sub-tasks. In this work, we present a universal instance perception model of the next generation, termed UNINEXT. UNINEXT reformulates diverse instance perception tasks into a unified object discovery and retrieval paradigm and can flexibly perceive different types of objects by simply changing the input prompts. This unified formulation brings the following benefits: (1) enormous data from different tasks and label vocabularies can be exploited for jointly training general instance-level representations, which is especially beneficial for tasks lacking in training data. (2) the unified model is parameter-efficient and can save redundant computation when handling multiple tasks simultaneously. UNINEXT shows superior performance on 20 challenging benchmarks from 10 instance-level tasks including classical image-level tasks (object detection and instance segmentation), vision-and-language tasks (referring expression comprehension and segmentation), and six video-level object tracking tasks. Code is available at https://github.com/MasterBin-IIAU/UNINEXT.

This paper aims to establish a generic multi-modal foundation model that has the scalable capability to massive downstream applications in E-commerce. Recently, large-scale vision-language pretraining approaches have achieved remarkable advances in the general domain. However, due to the significant differences between natural and product images, directly applying these frameworks for modeling image-level representations to E-commerce will be inevitably sub-optimal. To this end, we propose an instance-centric multi-modal pretraining paradigm called ECLIP in this work. In detail, we craft a decoder architecture that introduces a set of learnable instance queries to explicitly aggregate instance-level semantics. Moreover, to enable the model to focus on the desired product instance without reliance on expensive manual annotations, two specially configured pretext tasks are further proposed. Pretrained on the 100 million E-commerce-related data, ECLIP successfully extracts more generic, semantic-rich, and robust representations. Extensive experimental results show that, without further fine-tuning, ECLIP surpasses existing methods by a large margin on a broad range of downstream tasks, demonstrating the strong transferability to real-world E-commerce applications.

Learning with large-scale unlabeled data has become a powerful tool for pre-training Visual Transformers (VTs). However, prior works tend to overlook that, in real-world scenarios, the input data may be corrupted and unreliable. Pre-training VTs on such corrupted data can be challenging, especially when we pre-train via the masked autoencoding approach, where both the inputs and masked “ground truth” targets can potentially be unreliable in this case. To address this limitation, we introduce the Token Boosting Module (TBM) as a plug-and-play component for VTs that effectively allows the VT to learn to extract clean and robust features during masked autoencoding pre-training. We provide theoretical analysis to show how TBM improves model pre-training with more robust and generalizable representations, thus benefiting down stream tasks. We conduct extensive experiments to analyze TBM's effectiveness, and results on four corrupted datasets demonstrate that TBM consistently improves performance on downstream tasks.

Referring expression segmentation aims to segment an object described by a language expression from an image. Despite the recent progress on this task, existing models tackling this task may not be able to fully capture semantics and visual representations of individual concepts, which limits their generalization capability, especially when handling novel compositions of learned concepts. In this work, through the lens of meta learning, we propose a Meta Compositional Referring Expression Segmentation (MCRES) framework to enhance model compositional generalization performance. Specifically, to handle various levels of novel compositions, our framework first uses training data to construct a virtual training set and multiple virtual testing sets, where data samples in each virtual testing set contain a level of novel compositions w.r.t. the virtual training set. Then, following a novel meta optimization scheme to optimize the model to obtain good testing performance on the virtual testing sets after training on the virtual training set, our framework can effectively drive the model to better capture semantics and visual representations of individual concepts, and thus obtain robust generalization performance even when handling novel compositions. Extensive experiments on three benchmark datasets demonstrate the effectiveness of our framework.

Learning image classification and image generation using the same set of network parameters presents a formidable challenge. Recent advanced approaches perform well in one task often exhibit poor performance in the other. This work introduces an energy-based classifier and generator, namely EGC, which can achieve superior performance in both tasks using a single neural network. Unlike conventional classifiers that produce a label given an image (i.e., a conditional distribution p(y|x)), the forward pass in EGC is a classification model that yields a joint distribution p(x,y), enabling a diffusion model in its backward pass by marginalizing out the label y to estimate the score function. Furthermore, EGC can be adapted for unsupervised learning by considering the label as latent variables. EGC achieves competitive generation results compared with state-of-the-art approaches on ImageNet-1k, CelebA-HQ and LSUN Church, while achieving superior classification accuracy and robustness against adversarial attacks on CIFAR-10. This work marks the inaugural success in mastering both domains using a unified network parameter set. We believe that EGC bridges the gap between discriminative and generative learning. Code will be released at https://github.com/GuoQiushan/EGC.

The reference-based object segmentation tasks, namely referring image segmentation (RIS), referring video object segmentation (RVOS), and video object segmentation (VOS), aim to segment a specific object by utilizing either language or annotated masks as references. Despite significant progress in each respective field, current methods are task-specifically designed and developed in different directions, which hinders the activation of multi-task capabilities for these tasks. In this work, we end the current fragmented situation and propose UniRef to unify the three reference-based object segmentation tasks with a single architecture. At the heart of our approach is the multiway-fusion for handling different task with respect to their specified references. And a unified Transformer architecture is then adopted for performing instance-level segmentation. With the unified designs, UniRef can be jointly trained on a broad range of benchmarks and can flexibly perform multiple tasks at runtime by specifying the corresponding references. We evaluate the jointly trained network on various benchmarks. Extensive experimental results indicate that our proposed UniRef achieves state-of-the-art performance on RIS and RVOS, and performs competitively on VOS with a single network.

Open-world instance segmentation is a rising task, which aims to segment all objects in the image by learning from a limited number of base-category objects. This task is challenging, as the number of unseen categories could be hundreds of times larger than that of seen categories. Recently, the DETR-like models have been extensively studied in the closed world while stay unexplored in the open world. In this paper, we utilize the Transformer for open-world instance segmentation and present SWORD. Firstly, we introduce to attach the stop-gradient operation before classification head and further add IoU heads for discovering novel objects. We demonstrate that a simple stop-gradient operation not only prevents the novel objects from being suppressed as background, but also allows the network to enjoy the merit of heuristic label assignment. Secondly, we propose a novel contrastive learning framework to enlarge the representations between objects and background. Specifically, we maintain a universal object queue to obtain the object center, and dynamically select positive and negative samples from the object queries for contrastive learning. While the previous works only focus on pursuing average recall and neglect average precision, we show the prominence of SWORD by giving consideration to both criteria. Our models achieve state-of-the-art performance in various open-world cross-category and cross-dataset generalizations. Particularly, in VOC to non-VOC setup, our method sets new state-of-the-art results of 40.0% on ${\text{AR}}_{100}^{\text{b}}$ and 34.9% on ${\text{AR}}_{100}^{\text{m}}$. For COCO to UVO generalization, SWORD significantly outperforms the previous best open-world model by 5.9% on APmand 8.1% on ${\text{AR}}_{100}^{\text{m}}$.

Open-World Instance Segmentation (OWIS) is an emerging research topic that aims to segment class-agnostic object instances from images. The mainstream approaches use a two-stage segmentation framework, which first locates the candidate object bounding boxes and then performs instance segmentation. In this work, we instead promote a single-stage transformer-based framework for OWIS. We argue that the end-to-end training process in the single-stage framework can be more convenient for directly regularizing the localization of class-agnostic object pixels. Based on the transformer-based instance segmentation framework, we propose a regularization model to predict foreground pixels and use its relation to instance segmentation to construct a cross-task consistency loss. We show that such a consistency loss could alleviate the problem of incomplete instance annotation - a common problem in the existing OWIS datasets. We also show that the proposed loss lends itself to an effective solution to semi-supervised OWIS that could be considered an extreme case that all object annotations are absent for some images. Our extensive experiments demonstrate that the proposed method achieves impressive results in both fully-supervised and semi-supervised settings. Compared to SOTA methods, the proposed method significantly improves the AP_100 score by 4.75% in UVO dataset →UVO dataset setting and 4.05% in COCO dataset →UVO dataset setting.

Deriving reliable region-word alignment from image-text pairs is critical to learn object-level vision-language representations for open-vocabulary object detection. Existing methods typically rely on pre-trained or self-trained vision-language models for alignment, which are prone to limitations in localization accuracy or generalization capabilities. In this paper, we propose CoDet, a novel approach that overcomes the reliance on pre-aligned vision-language space by reformulating region-word alignment as a co-occurring object discovery problem. Intuitively, by grouping images that mention a shared concept in their captions, objects corresponding to the shared concept shall exhibit high co-occurrence among the group. CoDet then leverages visual similarities to discover the co-occurring objects and align them with the shared concept. Extensive experiments demonstrate that CoDet has superior performances and compelling scalability in open-vocabulary detection, e.g., by scaling up the visual backbone, CoDet achieves 37.0 $AP^m_{novel}$ and 44.7 $AP^m_{all}$ on OV-LVIS, surpassing the previous SoTA by 4.2 $AP^m_{novel}$ and 9.8 $AP^m_{all}$. Code is available at https://github.com/CVMI-Lab/CoDet.

Multimodal representation learning has shown promising improvements on various vision-language tasks (e.g., image-text retrieval, visual question answering, etc) and has significantly advanced the development of multimedia information systems. Most existing methods excel at building global-level alignment between vision and language while lacking effective fine-grained image-text interaction. In this paper, we propose a jointly masked multimodal modeling method to learn fine-grained multimodal representations. Our method performs joint masking on image-text input and integrates both implicit and explicit targets for the masked signals to recover. The implicit target provides a unified and debiased objective for vision and language, where the model predicts latent multimodal representations of the unmasked input. The explicit target further enriches the multimodal representations by recovering high-level and semantically meaningful information: momentum visual features of image patches and concepts of word tokens. Through such a masked modeling process, our model not only learns fine-grained multimodal interaction, but also avoids the semantic gap between high-level representations and low-or mid-level prediction targets (e.g., image pixels, discrete vision tokens), thus producing semantically rich multimodal representations that perform well on both zero-shot and fine-tuned settings. Our pre-trained model (named MAMO) achieves state-of-the-art performance on various downstream vision-language tasks, including image-text retrieval, visual question answering, visual reasoning, and weakly-supervised visual grounding.

Person search aims at localizing and recognizing query persons from raw video frames, which is a combination of two sub-tasks, i.e., pedestrian detection and person re-identification. The dominant fashion is termed as the one-step person search that jointly optimizes detection and identification in a unified network, exhibiting higher efficiency. However, there remain major challenges: (i) conflicting objectives of multiple sub-tasks under the shared feature space, (ii) inconsistent memory bank caused by the limited batch size, (iii) underutilized unlabeled identities during the identification learning. To address these issues, we develop an enhanced decoupled and memory-reinforced network (DMRNet++). First, we simplify the standard tightly coupled pipelines and establish a task-decoupled framework (TDF). Second, we build a memory-reinforced mechanism (MRM), with a slow-moving average of the network to better encode the consistency of the memorized features. Third, considering the potential of unlabeled samples, we model the recognition process as semi-supervised learning. An unlabeled-aided contrastive loss (UCL) is developed to boost the identification feature learning by exploiting the aggregation of unlabeled identities. Experimentally, the proposed DMRNet++ obtains the mAP of 94.5% and 52.1% on CUHK-SYSU and PRW datasets, which exceeds most existing methods.

Co-occurrent visual pattern makes context aggregation become an essential paradigm for semantic segmentation. The existing studies focus on modeling the contexts within image while neglecting the valuable semantics of the corresponding category beyond image. To this end, we propose a novel soft mining contextual information beyond image paradigm named MCIBI++ to further boost the pixel-level representations. Specifically, we first set up a dynamically updated memory module to store the dataset-level distribution information of various categories and then leverage the information to yield the dataset-level category representations during network forward. After that, we generate a class probability distribution for each pixel representation and conduct the dataset-level context aggregation with the class probability distribution as weights. Finally, the original pixel representations are augmented with the aggregated dataset-level and the conventional image-level contextual information. Moreover, in the inference phase, we additionally design a coarse-to-fine iterative inference strategy to further boost the segmentation results. MCIBI++ can be effortlessly incorporated into the existing segmentation frameworks and bring consistent performance improvements. Also, MCIBI++ can be extended into the video semantic segmentation framework with considerable improvements over the baseline. Equipped with MCIBI++, we achieved the state-of-the-art performance on seven challenging image or video semantic segmentation benchmarks.

Object detection serves as one of most fundamental computer vision tasks. Existing works on object detection heavily rely on dense object candidates, such as k anchor boxes pre-defined on all grids of an image feature map of size H×W. In this paper, we present Sparse R-CNN, a very simple and sparse method for object detection in images. In our method, a fixed sparse set of learned object proposals ( N in total) are provided to the object recognition head to perform classification and localization. By replacing HWk (up to hundreds of thousands) hand-designed object candidates with N (e.g., 100) learnable proposals, Sparse R-CNN makes all efforts related to object candidates design and one-to-many label assignment completely obsolete. More importantly, Sparse R-CNN directly outputs predictions without the non-maximum suppression (NMS) post-processing procedure. Thus, it establishes an end-to-end object detection framework. Sparse R-CNN demonstrates highly competitive accuracy, run-time and training convergence performance with the well-established detector baselines on the challenging COCO dataset and CrowdHuman dataset. We hope that our work can inspire re-thinking the convention of dense prior in object detectors and designing new high-performance detectors.

Generally, humans are more skilled at perceiving differences between high-quality (HQ) and low-quality (LQ) images than directly judging the quality of a single LQ image. This situation also applies to image quality assessment (IQA). Although recent no-reference (NR-IQA) methods have made great progress to predict image quality free from the reference image, they still have the potential to achieve better performance since HQ image information is not fully exploited. In contrast, full-reference (FR-IQA) methods tend to provide more reliable quality evaluation, but its practicability is affected by the requirement for pixel-level aligned reference images. To address this, we firstly propose the content-variant reference method via knowledge distillation (CVRKD-IQA). Specifically, we use non-aligned reference (NAR) images to introduce various prior distributions of high-quality images. The comparisons of distribution differences between HQ and LQ images can help our model better assess the image quality. Further, the knowledge distillation transfers more HQ-LQ distribution difference information from the FR-teacher to the NAR-student and stabilizing CVRKD-IQA performance. Moreover, to fully mine the local-global combined information, while achieving faster inference speed, our model directly processes multiple image patches from the input with the MLP-mixer. Cross-dataset experiments verify that our model can outperform all NAR/NR-IQA SOTAs, even reach comparable performance than FR-IQA methods on some occasions. Since the content-variant and non-aligned reference HQ images are easy to obtain, our model can support more IQA applications with its robustness to content variations. Our code is available: https://github.com/guanghaoyin/CVRKD-IQA.

A typical pipeline for multi-object tracking (MOT) is to use a detector for object localization, and following re-identification (re-ID)for object association. This pipeline is partially motivated by recent progress in both object detection and re- ID, and partially motivated by biases in existing tracking datasets, where most objects tend to have distin-guishing appearance and re-ID models are sufficient for es-tablishing associations. In response to such bias, we would like to re-emphasize that methods for multi-object tracking should also work when object appearance is not sufficiently discriminative. To this end, we propose a large-scale dataset for multi-human tracking, where humans have sim-ilar appearance, diverse motion and extreme articulation. As the dataset contains mostly group dancing videos, we name it “DanceTrack”. We expect DanceTrack to provide a better platform to develop more MOT algorithms that rely less on visual discrimination and depend more on motion analysis. We benchmark several state-of-the-art trackers on our dataset and observe a significant performance drop on DanceTrack when compared against existing benchmarks. The dataset, project code and competition is released at: https://github.com/DanceTrack.

Referring video object segmentation (R-VOS) is an emerging cross-modal task that aims to segment the target object referred by a language expression in all video frames. In this work, we propose a simple and unified framework built upon Transformer, termed ReferFormer. It views the language as queries and directly attends to the most relevant regions in the video frames. Concretely, we introduce a small set of object queries conditioned on the language as the input to the Transformer. In this manner, all the queries are obligated to find the referred objects only. They are eventually transformed into dynamic kernels which capture the crucial object-level information, and play the role of convolution filters to generate the segmentation masks from feature maps. The object tracking is achieved naturally by linking the corresponding queries across frames. This mechanism greatly simplifies the pipeline and the end-to-end framework is significantly different from the previous methods. Extensive experiments on Ref-Youtube-vos, Ref-DAVIS17, A2D-Sentences and JHMDB-Sentences show the effectiveness of ReferFormer. On Ref-Youtube-vos, ReferFormer achieves 55.6 J&F with a ResNet-50 backbone without bells and whistles, which exceeds the previous state-of-the-art performance by 8.4 points. In addition, with the strong Video-Swin-Base backbone, ReferFormer achieves the best J&F of 64.9 among all existing methods. Moreover, we show the impressive results of 55.0 mAP and 43.7 mAP on A2D-Sentences and JHMDB-Sentences respectively, which significantly outperforms the previous methods by a large margin. Code is publicly available at https://github.com/wjn922/ReferFormer.

Knowledge distillation has been applied to image classification successfully. However, object detection is much more sophisticated and most knowledge distillation methods have failed on it. In this paper, we point out that in object detection, the features of the teacher and student vary greatly in different areas, especially in the foreground and background. If we distill them equally, the uneven differences between feature maps will negatively affect the distillation. Thus, we propose Focal and Global Distillation (FGD). Focal distillation separates the foreground and background, forcing the student to focus on the teacher's critical pixels and channels. Global distillation rebuilds the relation between different pixels and transfers it from teachers to students, compensating for missing global information in focal distillation. As our method only needs to calculate the loss on the feature map, FGD can be applied to various detectors. We experiment on various detectors with different backbones and the results show that the student detector achieves excellent mAP improvement. For example, ResNet-50 based RetinaNet, Faster RCNN, RepPoints and Mask RCNN with our distillation method achieve 40.7%, 42.0%, 42.0% and 42.1% mAP on COCO2017, which are 3.3, 3.6, 3.4 and 2.9 higher than the baseline, respectively. Our codes are available at https://github.com/yzd-v/FGD.

Spatio-Temporal video grounding (STVG) focuses on retrieving the spatio-temporal tube of a specific object depicted by a free-form textual expression. Existing approaches mainly treat this complicated task as a parallel frame-grounding problem and thus suffer from two types of inconsistency drawbacks: feature alignment inconsistency and prediction inconsistency. In this paper, we present an end-to-end one-stage framework, termed Spatio-Temporal Consistency-Aware Transformer (STCAT), to alleviate these issues. Specially, we introduce a novel multi-modal template as the global objective to address this task, which explicitly constricts the grounding region and associates the predictions among all video frames. Moreover, to generate the above template under sufficient video-textual perception, an encoder-decoder architecture is proposed for effective global context modeling. Thanks to these critical designs, STCAT enjoys more consistent cross-modal feature alignment and tube prediction without reliance on any pre-trained object detectors. Extensive experiments show that our method outperforms previous state-of-the-arts with clear margins on two challenging video benchmarks (VidSTG and HC-STVG), illustrating the superiority of the proposed framework to better understanding the association between vision and natural language. Code is publicly available at https://github.com/jy0205/STCAT.

In this paper, we empirically study how to make the most of low-resolution frames for efficient video recognition. Existing methods mainly focus on developing compact networks or alleviating temporal redundancy of video inputs to increase efficiency, whereas compressing frame resolution has rarely been considered a promising solution. A major concern is the poor recognition accuracy on low-resolution frames. We thus start by analyzing the underlying causes of performance degradation on low-resolution frames. Our key finding is that the major cause of degradation is not information loss in the down-sampling process, but rather the mismatch between network architecture and input scale. Motivated by the success of knowledge distillation (KD), we propose to bridge the gap between network and input size via cross-resolution KD (ResKD). Our work shows that ResKD is a simple but effective method to boost recognition accuracy on low-resolution frames. Without bells and whistles, ResKD considerably surpasses all competitive methods in terms of efficiency and accuracy on four large-scale benchmark datasets, i.e., ActivityNet, FCVID, Mini-Kinetics, Something-Something V2. In addition, we extensively demonstrate its effectiveness over state-of-the-art architectures, i.e., 3D-CNNs and Video Transformers, and scalability towards super low-resolution frames. The results suggest ResKD can serve as a general inference acceleration method for state-of-the-art video recognition. Our code will be available at https://github.com/CVMI-Lab/ResKD.

We propose a sparse end-to-end multi-person pose regression framework, termed QueryPose, which can directly predict multi-person keypoint sequences from the input image. The existing end-to-end methods rely on dense representations to preserve the spatial detail and structure for precise keypoint localization. However, the dense paradigm introduces complex and redundant post-processes during inference. In our framework, each human instance is encoded by several learnable spatial-aware part-level queries associated with an instance-level query. First, we propose the Spatial Part Embedding Generation Module (SPEGM) that considers the local spatial attention mechanism to generate several spatial-sensitive part embeddings, which contain spatial details and structural information for enhancing the part-level queries. Second, we introduce the Selective Iteration Module (SIM) to adaptively update the sparse part-level queries via the generated spatial-sensitive part embeddings stage-by-stage. Based on the two proposed modules, the part-level queries are able to fully encode the spatial details and structural information for precise keypoint regression. With the bipartite matching, QueryPose avoids the hand-designed post-processes. Without bells and whistles, QueryPose surpasses the existing dense end-to-end methods with 73.6 AP on MS COCO mini-val set and 72.7 AP on CrowdPose test set. Code is available at https://github.com/buptxyb666/QueryPose.

Although the single-image super-resolution (SISR) methods have achieved great success on the single degradation, they still suffer performance drop with multiple degrading effects in real scenarios. Recently, some blind and non-blind models for multiple degradations have been explored. However, these methods usually degrade significantly for distribution shifts between the training and test data. Towards this end, we propose a novel conditional hyper-network framework for super-resolution with multiple degradations (named CMDSR), which helps the SR framework learn how to adapt to changes in the degradation distribution of input. We extract degradation prior at the task-level with the proposed ConditionNet, which will be used to adapt the parameters of the basic SR network (BaseNet). Specifically, the ConditionNet of our framework first learns the degradation prior from a support set, which is composed of a series of degraded image patches from the same task. Then the adaptive BaseNet rapidly shifts its parameters according to the conditional features. Moreover, in order to better extract degradation prior, we propose a task contrastive loss to shorten the inner-task distance and enlarge the cross-task distance between task-level features. Without predefining degradation maps, our blind framework can conduct one single parameter update to yield considerable improvement in SR results. Extensive experiments demonstrate the effectiveness of CMDSR over various blind, and even several non-blind methods. The flexible BaseNet structure also reveals that CMDSR can be a general framework for a large series of SISR models. Our code is available at https://github.com/guanghaoyin/CMDSR.

Unsupervised domain adaptation (UDA) aims to enhance the generalization capability of a certain model from a source domain to a target domain. Present UDA models focus on alleviating the domain shift by minimizing the feature discrepancy between the source domain and the target domain but usually ignore the class confusion problem. In this work, we propose an Inter-class Separation and Intra-class Aggregation (ISIA) mechanism. It encourages the cross-domain representative consistency between the same categories and differentiation among diverse categories. In this way, the features belonging to the same categories are aligned together and the confusable categories are separated. By measuring the align complexity of each category, we design an Adaptive-weighted Instance Matching (AIM) strategy to further optimize the instance-level adaptation. Based on our proposed methods, we also raise a hierarchical unsupervised domain adaptation framework for cross-domain semantic segmentation task. Through performing the image-level, feature-level, category-level and instance-level alignment, our method achieves a stronger generalization performance of the model from the source domain to the target domain. In two typical cross-domain semantic segmentation tasks, i.e., GTA 5→ Cityscapes and SYNTHIA → Cityscapes, our method achieves the state-of-the-art segmentation accuracy. We also build two cross-domain semantic segmentation datasets based on the publicly available data, i.e., remote sensing building segmentation and road segmentation, for domain adaptive segmentation. Our code, models and datasets are available at https://github.com/HibiscusYB/BAFFT.

Generative adversarial networks (GANs) have achieved remarkable progress in recent years, but the continuously growing scale of models make them challenging to deploy widely in practical applications. In particular, for real-time generation tasks, different devices require generators of different sizes due to varying computing power. In this paper, we introduce slimmable GANs (SlimGANs), which can flexibly switch the width of the generator to accommodate various quality-efficiency trade-offs at runtime. Specifically, we leverage multiple discriminators that share partial parameters to train the slimmable generator. To facilitate the consistency between generators of different widths, we present a stepwise inplace distillation technique that encourages narrow generators to learn from wide ones. As for class-conditional generation, we propose a sliceable conditional batch normalization that incorporates the label information into different widths. Our methods are validated, both quantitatively and qualitatively, by extensive experiments and a detailed ablation study.

We present Sparse R-CNN, a purely sparse method for object detection in images. Existing works on object detection heavily rely on dense object candidates, such as k anchor boxes pre-defined on all grids of image feature map of size H × W. In our method, however, a fixed sparse set of learned object proposals, total length of N, are provided to object recognition head to perform classification and location. By eliminating HWk (up to hundreds of thousands) hand-designed object candidates to N (e.g. 100) learnable proposals, Sparse R-CNN completely avoids all efforts related to object candidates design and many-to-one label assignment. More importantly, final predictions are directly output without non-maximum suppression post-procedure. Sparse R-CNN demonstrates accuracy, run-time and training convergence performance on par with the well-established detector baselines on the challenging COCO dataset, e.g., achieving 45.0 AP in standard 3× training schedule and running at 22 fps using ResNet-50 FPN model. We hope our work could inspire re-thinking the convention of dense prior in object detectors. The code is available at: https://github.com/PeizeSun/SparseR-CNN.

Supervised learning is dominant in person search, but it requires elaborate labeling of bounding boxes and identities. Large-scale labeled training data is often difficult to collect, especially for person identities. A natural question is whether a good person search model can be trained without the need of identity supervision. In this paper, we present a weakly supervised setting where only bounding box annotations are available. Based on this new setting, we provide an effective baseline model termed Region Siamese Networks (R-SiamNets). Towards learning useful representations for recognition in the absence of identity labels, we supervise the R-SiamNet with instance-level consistency loss and cluster-level contrastive loss. For instance-level consistency learning, the R-SiamNet is constrained to extract consistent features from each person region with or without out-of-region context. For cluster-level contrastive learning, we enforce the aggregation of closest instances and the separation of dissimilar ones in feature space. Extensive experiments validate the utility of our weakly supervised method. Our model achieves the rank-1 of 87.1% and mAP of 86.0% on CUHK-SYSU benchmark, which surpasses several fully supervised methods, such as OIM [36] and MGTS [4], by a clear margin. More promising performance can be reached by incorporating extra training data. We hope this work could encourage the future research in this field.

We address the problem of domain generalizable object detection, which aims to learn a domain-invariant detector from multiple "seen" domains so that it can generalize well to other "unseen" domains. The generalization ability is crucial in practical scenarios especially when it is difficult to collect data. Compared to image classification, domain generalization in object detection has seldom been explored with more challenges brought by domain gaps on both image and instance levels. In this paper, we propose a novel generalizable object detection model, termed Domain-Invariant Disentangled Network (DIDN). In contrast to directly aligning multiple sources, we integrate a disentangled network into Faster R-CNN. By disentangling representations on both image and instance levels, DIDN is able to learn domain-invariant representations that are suitable for generalized object detection. Furthermore, we design a cross-level representation reconstruction to complement this two-level disentanglement so that informative object representations could be preserved. Extensive experiments are conducted on five benchmark datasets and the results demonstrate that our model achieves state-of-the-art performances on domain generalization for object detection.

Most existing convolution neural network (CNN) based super-resolution (SR) methods generate their paired training dataset by artificially synthesizing low-resolution (LR) images from the high-resolution (HR) ones. However, this dataset preparation strategy harms the application of these CNNs in real-world scenarios due to the inherent domain gap between the training and testing data. A popular attempts towards the challenge is unpaired generative adversarial networks, which generate "real" LR counterparts from real HR images using image-to-image translation and then perform super-resolution from "real" LR→SR. Despite great progress, it is still difficult to synthesize perfect "real" LR images for super-resolution. In this paper, we firstly consider the real-world SR problem from the traditional domain adaptation perspective. We propose a novel unpaired SR training framework based on feature distribution alignment, with which we can obtain degradation-indistinguishable feature maps and then map them to HR images. In order to generate better SR images for target LR domain, we introduce several regularization losses to force the aligned feature to locate around the target domain. Our experiments indicate that our SR network obtains the state-of-the-art performance over both blind and unpaired SR methods on diverse datasets.

Object detection has recently achieved a breakthrough for removing the last one non-differentiable component in the pipeline, Non-Maximum Suppression (NMS), and building up an end-to-end system. However, what makes for its one-to-one prediction has not been well understood. In this paper, we first point out that one-to-one positive sample assignment is the key factor, while, one-to-many assignment in previous detectors causes redundant predictions in inference. Second, we surprisingly find that even training with one-to-one assignment, previous detectors still produce redundant predictions. We identify that classification cost in matching cost is the main ingredient: (1) previous detectors only consider location cost, (2) by additionally introducing classification cost, previous detectors immediately produce one-to-one prediction during inference. We introduce the concept of score gap to explore the effect of matching cost. Classification cost enlarges the score gap by choosing positive samples as those of highest score in the training iteration and reducing noisy positive samples brought by only location cost. Finally, we demonstrate the advantages of end-to-end object detection on crowded scenes.

With the growing number of videos in video sharing platforms, how to facilitate the searching and browsing of the user-generated video has attracted intense attention by multimedia community. To help people efficiently search and browse relevant videos, summaries of videos become important. The prior works in multimodal video summarization mainly explore visual and ASR tokens as two separate sources and struggle to fuse the multimodal information for generating the summaries. However, the time information inside videos is commonly ignored. In this paper, we find that it is important to leverage the timestamps to accurately incorporate multimodal signals for the task. We propose a Time-Aware Multimodal Transformer (TAMT) with a novel short-term order-sensitive attention mechanism. The attention mechanism can attend the inputs differently based on time difference to explore the time information inherent inside video more thoroughly. As such, TAMT can fuse the different modalities better for summarizing the videos. Experiments show that our proposed approach is effective and achieves the state-of-the-art performances on both YouCookII and open-domain How2 datasets.

Recently, self-supervised learning for graph neural networks (GNNs) has attracted considerable attention because of their notable successes in learning the representation of graph-structure data. However, the formation of a real-world graph typically arises from the highly complex interaction of many latent factors. The existing self-supervised learning methods for GNNs are inherently holistic and neglect the entanglement of the latent factors, resulting in the learned representations suboptimal for downstream tasks and difficult to be interpreted. Learning disentangled graph representations with self-supervised learning poses great challenges and remains largely ignored by the existing literature. In this paper, we introduce the Disentangled Graph Contrastive Learning (DGCL) method, which is able to learn disentangled graph-level representations with self-supervision. In particular, we first identify the latent factors of the input graph and derive its factorized representations. Each of the factorized representations describes a latent and disentangled aspect pertinent to a specific latent factor of the graph. Then we propose a novel factor-wise discrimination objective in a contrastive learning manner, which can force the factorized representations to independently reflect the expressive information from different latent factors. Extensive experiments on both synthetic and real-world datasets demonstrate the superiority of our method against several state-of-the-art baselines.

Modeling spatial or temporal long-range dependency plays a key role in deep neural networks. Conventional dominant solutions include recurrent operations on sequential data or deeply stacking convolutional layers with small kernel size. Recently, a number of non-local operators (such as self-attention based) have been devised. They are typically generic and can be plugged into many existing network pipelines for globally computing among any two neurons in a feature map. This work proposes a novel non-local operator. It is inspired by the attention mechanism of human visual system, which can quickly attend to important local parts in sight and suppress other less-relevant information. The core of our method is learnable and data-adaptive bilinear attentional transform (BA-Transform), whose merits are three-folds: first, BA-Transform is versatile to model a wide spectrum of local or global attentional operations, such as emphasizing specific local regions. Each BA-Transform is learned in a data-adaptive way; Secondly, to address the discrepancy among features, we further design grouped BA-Transforms, which essentially apply different attentional operations to different groups of feature channels; Thirdly, many existing non-local operators are computation-intensive. The proposed BA-Transform is implemented by simple matrix multiplication and admits better efficacy. For empirical evaluation, we perform comprehensive experiments on two large-scale benchmarks, ImageNet and Kinetics, for image / video classification respectively. The achieved accuracies and various ablation experiments consistently demonstrate significant improvement by large margins.

Orthogonality is widely used for training deep neural networks (DNNs) due to its ability to maintain all singular values of the Jacobian close to 1 and reduce redundancy in representation. This paper proposes a computationally efficient and numerically stable orthogonalization method using Newton's iteration (ONI), to learn a layer-wise orthogonal weight matrix in DNNs. ONI works by iteratively stretching the singular values of a weight matrix towards 1. This property enables it to control the orthogonality of a weight matrix by its number of iterations. We show that our method improves the performance of image classification networks by effectively controlling the orthogonality to provide an optimal tradeoff between optimization benefits and representational capacity reduction. We also show that ONI stabilizes the training of generative adversarial networks (GANs) by maintaining the Lipschitz continuity of a network, similar to spectral normalization (SN), and further outperforms SN by providing controllable orthogonality.

In recent years, video generation has attracted a lot of attention in the computer vision community. Unlike image generation which only focuses on appearance, video generation requires modeling both content information and motion dynamics. In this work, we propose MoFlowGAN, which explicitly models motion dynamics by a content-motion decomposition architecture with an additional flow generator. The decomposition architecture models content and motion separately and is instantiated by a compact variant of BigGAN [1]. Besides, the flow generator generates optical flow directly based on high-level feature maps of adjacent frames as a strong supervision, hence the searching space of motion patterns is highly reduced. Our proposed MoFlowGAN achieves the state-of-the-art results on both MUG facial expression and UCF-101 datasets.