Jianmin Bao

Abdelrahman Abouelenin, Atabak Ashfaq, Adam Atkinson et al. · microsoft-research

We introduce Phi-4-Mini and Phi-4-Multimodal, compact yet highly capable language and multimodal models. Phi-4-Mini is a 3.8-billion-parameter language model trained on high-quality web and synthetic data, significantly outperforming recent open-source models of similar size and matching the performance of models twice its size on math and coding tasks requiring complex reasoning. This achievement is driven by a carefully curated synthetic data recipe emphasizing high-quality math and coding datasets. Compared to its predecessor, Phi-3.5-Mini, Phi-4-Mini features an expanded vocabulary size of 200K tokens to better support multilingual applications, as well as group query attention for more efficient long-sequence generation. Phi-4-Multimodal is a multimodal model that integrates text, vision, and speech/audio input modalities into a single model. Its novel modality extension approach leverages LoRA adapters and modality-specific routers to allow multiple inference modes combining various modalities without interference. For example, it now ranks first in the OpenASR leaderboard to date, although the LoRA component of the speech/audio modality has just 460 million parameters. Phi-4-Multimodal supports scenarios involving (vision + language), (vision + speech), and (speech/audio) inputs, outperforming larger vision-language and speech-language models on a wide range of tasks. Additionally, we experiment to further train Phi-4-Mini to enhance its reasoning capabilities. Despite its compact 3.8-billion-parameter size, this experimental version achieves reasoning performance on par with or surpassing significantly larger models, including DeepSeek-R1-Distill-Qwen-7B and DeepSeek-R1-Distill-Llama-8B.

Tengfei Wang, Bo Zhang, Ting Zhang et al. · microsoft-research

This paper presents a 3D generative model that uses diffusion models to automatically generate 3D digital avatars represented as neural radiance fields. A significant challenge in generating such avatars is that the memory and processing costs in 3D are prohibitive for producing the rich details required for high-quality avatars. To tackle this problem we propose the roll-out diffusion network (Rodin), which represents a neural radiance field as multiple 2D feature maps and rolls out these maps into a single 2D feature plane within which we perform 3D-aware diffusion. The Rodin model brings the much-needed computational efficiency while preserving the integrity of diffusion in 3D by using 3D-aware convolution that attends to projected features in the 2D feature plane according to their original relationship in 3D. We also use latent conditioning to orchestrate the feature generation for global coherence, leading to high-fidelity avatars and enabling their semantic editing based on text prompts. Finally, we use hierarchical synthesis to further enhance details. The 3D avatars generated by our model compare favorably with those produced by existing generative techniques. We can generate highly detailed avatars with realistic hairstyles and facial hair like beards. We also demonstrate 3D avatar generation from image or text as well as text-guided editability.

Zhendong Wang, Jianmin Bao, Wengang Zhou et al.

Diffusion models have shown remarkable success in visual synthesis, but have also raised concerns about potential abuse for malicious purposes. In this paper, we seek to build a detector for telling apart real images from diffusion-generated images. We find that existing detectors struggle to detect images generated by diffusion models, even if we include generated images from a specific diffusion model in their training data. To address this issue, we propose a novel image representation called DIffusion Reconstruction Error (DIRE), which measures the error between an input image and its reconstruction counterpart by a pre-trained diffusion model. We observe that diffusion-generated images can be approximately reconstructed by a diffusion model while real images cannot. It provides a hint that DIRE can serve as a bridge to distinguish generated and real images. DIRE provides an effective way to detect images generated by most diffusion models, and it is general for detecting generated images from unseen diffusion models and robust to various perturbations. Furthermore, we establish a comprehensive diffusion-generated benchmark including images generated by eight diffusion models to evaluate the performance of diffusion-generated image detectors. Extensive experiments on our collected benchmark demonstrate that DIRE exhibits superiority over previous generated-image detectors. The code and dataset are available at https://github.com/ZhendongWang6/DIRE.

Xiaoyi Dong, Jianmin Bao, Yinglin Zheng et al.

This paper presents a simple yet effective framework MaskCLIP, which incorporates a newly proposed masked self-distillation into contrastive language-image pretraining. The core idea of masked self-distillation is to distill representation from a full image to the representation predicted from a masked image. Such incorporation enjoys two vital benefits. First, masked self-distillation targets local patch representation learning, which is complementary to vision-language contrastive focusing on text-related representation. Second, masked self-distillation is also consistent with vision-language contrastive from the perspective of training objective as both utilize the visual encoder for feature aligning, and thus is able to learn local semantics getting indirect supervision from the language. We provide specially designed experiments with a comprehensive analysis to validate the two benefits. Symmetrically, we also introduce the local semantic supervision into the text branch, which further improves the pretraining performance. With extensive experiments, we show that MaskCLIP, when applied to various challenging downstream tasks, achieves superior results in linear probing, finetuning, and zero-shot performance with the guidance of the language encoder. Code will be release at \url{https://github.com/LightDXY/MaskCLIP}.

Tiankai Hang, Shuyang Gu, Chen Li et al.

Denoising diffusion models have been a mainstream approach for image generation, however, training these models often suffers from slow convergence. In this paper, we discovered that the slow convergence is partly due to conflicting optimization directions between timesteps. To address this issue, we treat the diffusion training as a multi-task learning problem, and introduce a simple yet effective approach referred to as Min-SNR-$γ$. This method adapts loss weights of timesteps based on clamped signal-to-noise ratios, which effectively balances the conflicts among timesteps. Our results demonstrate a significant improvement in converging speed, 3.4$\times$ faster than previous weighting strategies. It is also more effective, achieving a new record FID score of 2.06 on the ImageNet $256\times256$ benchmark using smaller architectures than that employed in previous state-of-the-art. The code is available at https://github.com/TiankaiHang/Min-SNR-Diffusion-Training.

Wengang Zhou, Weilun Wang, Jianmin Bao et al.

Denoising Diffusion Probabilistic Models (DDPMs) have achieved remarkable success in various image generation tasks compared with Generative Adversarial Nets (GANs). Recent work on semantic image synthesis mainly follows the de facto GAN-based approaches, which may lead to unsatisfactory quality or diversity of generated images. In this paper, we propose a novel framework based on DDPM for semantic image synthesis. Unlike previous conditional diffusion model directly feeds the semantic layout and noisy image as input to a U-Net structure, which may not fully leverage the information in the input semantic mask, our framework processes semantic layout and noisy image differently. It feeds noisy image to the encoder of the U-Net structure while the semantic layout to the decoder by multi-layer spatially-adaptive normalization operators. To further improve the generation quality and semantic interpretability in semantic image synthesis, we introduce the classifier-free guidance sampling strategy, which acknowledge the scores of an unconditional model for sampling process. Extensive experiments on four benchmark datasets demonstrate the effectiveness of our proposed method, achieving state-of-the-art performance in terms of fidelity (FID) and diversity (LPIPS). Our code and pretrained models are available at https://github.com/WeilunWang/semantic-diffusion-model.

Xiaoyi Dong, Jianmin Bao, Dongdong Chen et al.

In this work we propose Identity Consistency Transformer, a novel face forgery detection method that focuses on high-level semantics, specifically identity information, and detecting a suspect face by finding identity inconsistency in inner and outer face regions. The Identity Consistency Transformer incorporates a consistency loss for identity consistency determination. We show that Identity Consistency Transformer exhibits superior generalization ability not only across different datasets but also across various types of image degradation forms found in real-world applications including deepfake videos. The Identity Consistency Transformer can be easily enhanced with additional identity information when such information is available, and for this reason it is especially well-suited for detecting face forgeries involving celebrities. Code will be released at \url{https://github.com/LightDXY/ICT_DeepFake}

Yixuan Wei, Han Hu, Zhenda Xie et al.

Masked image modeling (MIM) learns representations with remarkably good fine-tuning performances, overshadowing previous prevalent pre-training approaches such as image classification, instance contrastive learning, and image-text alignment. In this paper, we show that the inferior fine-tuning performance of these pre-training approaches can be significantly improved by a simple post-processing in the form of feature distillation (FD). The feature distillation converts the old representations to new representations that have a few desirable properties just like those representations produced by MIM. These properties, which we aggregately refer to as optimization friendliness, are identified and analyzed by a set of attention- and optimization-related diagnosis tools. With these properties, the new representations show strong fine-tuning performance. Specifically, the contrastive self-supervised learning methods are made as competitive in fine-tuning as the state-of-the-art masked image modeling (MIM) algorithms. The CLIP models' fine-tuning performance is also significantly improved, with a CLIP ViT-L model reaching 89.0% top-1 accuracy on ImageNet-1K classification. On the 3-billion-parameter SwinV2-G model, the fine-tuning accuracy is improved by +1.5 mIoU / +1.1 mAP to 61.4 mIoU / 64.2 mAP on ADE20K semantic segmentation and COCO object detection, respectively, creating new records on both benchmarks. More importantly, our work provides a way for the future research to focus more effort on the generality and scalability of the learnt representations without being pre-occupied with optimization friendliness since it can be enhanced rather easily. The code will be available at https://github.com/SwinTransformer/Feature-Distillation.

Xiaoyi Dong, Jianmin Bao, Ting Zhang et al.

We propose bootstrapped masked autoencoders (BootMAE), a new approach for vision BERT pretraining. BootMAE improves the original masked autoencoders (MAE) with two core designs: 1) momentum encoder that provides online feature as extra BERT prediction targets; 2) target-aware decoder that tries to reduce the pressure on the encoder to memorize target-specific information in BERT pretraining. The first design is motivated by the observation that using a pretrained MAE to extract the features as the BERT prediction target for masked tokens can achieve better pretraining performance. Therefore, we add a momentum encoder in parallel with the original MAE encoder, which bootstraps the pretraining performance by using its own representation as the BERT prediction target. In the second design, we introduce target-specific information (e.g., pixel values of unmasked patches) from the encoder directly to the decoder to reduce the pressure on the encoder of memorizing the target-specific information. Thus, the encoder focuses on semantic modeling, which is the goal of BERT pretraining, and does not need to waste its capacity in memorizing the information of unmasked tokens related to the prediction target. Through extensive experiments, our BootMAE achieves $84.2\%$ Top-1 accuracy on ImageNet-1K with ViT-B backbone, outperforming MAE by $+0.8\%$ under the same pre-training epochs. BootMAE also gets $+1.0$ mIoU improvements on semantic segmentation on ADE20K and $+1.3$ box AP, $+1.4$ mask AP improvement on object detection and segmentation on COCO dataset. Code is released at https://github.com/LightDXY/BootMAE.

Zhendong Wang, Jianmin Bao, Wengang Zhou et al.

Existing face forgery detection models try to discriminate fake images by detecting only spatial artifacts (e.g., generative artifacts, blending) or mainly temporal artifacts (e.g., flickering, discontinuity). They may experience significant performance degradation when facing out-domain artifacts. In this paper, we propose to capture both spatial and temporal artifacts in one model for face forgery detection. A simple idea is to leverage a spatiotemporal model (3D ConvNet). However, we find that it may easily rely on one type of artifact and ignore the other. To address this issue, we present a novel training strategy called AltFreezing for more general face forgery detection. The AltFreezing aims to encourage the model to detect both spatial and temporal artifacts. It divides the weights of a spatiotemporal network into two groups: spatial-related and temporal-related. Then the two groups of weights are alternately frozen during the training process so that the model can learn spatial and temporal features to distinguish real or fake videos. Furthermore, we introduce various video-level data augmentation methods to improve the generalization capability of the forgery detection model. Extensive experiments show that our framework outperforms existing methods in terms of generalization to unseen manipulations and datasets. Code is available at https: //github.com/ZhendongWang6/AltFreezing.

Xiaoyi Dong, Jianmin Bao, Ting Zhang et al.

Recent studies have shown that CLIP has achieved remarkable success in performing zero-shot inference while its fine-tuning performance is not satisfactory. In this paper, we identify that fine-tuning performance is significantly impacted by hyper-parameter choices. We examine various key hyper-parameters and empirically evaluate their impact in fine-tuning CLIP for classification tasks through a comprehensive study. We find that the fine-tuning performance of CLIP is substantially underestimated. Equipped with hyper-parameter refinement, we demonstrate CLIP itself is better or at least competitive in fine-tuning compared with large-scale supervised pre-training approaches or latest works that use CLIP as prediction targets in Masked Image Modeling. Specifically, CLIP ViT-Base/16 and CLIP ViT-Large/14 can achieve 85.7%,88.0% finetuning Top-1 accuracy on the ImageNet-1K dataset . These observations challenge the conventional conclusion that CLIP is not suitable for fine-tuning, and motivate us to rethink recently proposed improvements based on CLIP. We will release our code publicly at \url{https://github.com/LightDXY/FT-CLIP}.

Hanqing Zhao, Dianmo Sheng, Jianmin Bao et al.

Copy-Paste is a simple and effective data augmentation strategy for instance segmentation. By randomly pasting object instances onto new background images, it creates new training data for free and significantly boosts the segmentation performance, especially for rare object categories. Although diverse, high-quality object instances used in Copy-Paste result in more performance gain, previous works utilize object instances either from human-annotated instance segmentation datasets or rendered from 3D object models, and both approaches are too expensive to scale up to obtain good diversity. In this paper, we revisit Copy-Paste at scale with the power of newly emerged zero-shot recognition models (e.g., CLIP) and text2image models (e.g., StableDiffusion). We demonstrate for the first time that using a text2image model to generate images or zero-shot recognition model to filter noisily crawled images for different object categories is a feasible way to make Copy-Paste truly scalable. To make such success happen, we design a data acquisition and processing framework, dubbed ``X-Paste", upon which a systematic study is conducted. On the LVIS dataset, X-Paste provides impressive improvements over the strong baseline CenterNet2 with Swin-L as the backbone. Specifically, it archives +2.6 box AP and +2.1 mask AP gains on all classes and even more significant gains with +6.8 box AP, +6.5 mask AP on long-tail classes. Our code and models are available at https://github.com/yoctta/XPaste.

Zixin Zhu, Xuelu Feng, Dongdong Chen et al.

StableDiffusion is a revolutionary text-to-image generator that is causing a stir in the world of image generation and editing. Unlike traditional methods that learn a diffusion model in pixel space, StableDiffusion learns a diffusion model in the latent space via a VQGAN, ensuring both efficiency and quality. It not only supports image generation tasks, but also enables image editing for real images, such as image inpainting and local editing. However, we have observed that the vanilla VQGAN used in StableDiffusion leads to significant information loss, causing distortion artifacts even in non-edited image regions. To this end, we propose a new asymmetric VQGAN with two simple designs. Firstly, in addition to the input from the encoder, the decoder contains a conditional branch that incorporates information from task-specific priors, such as the unmasked image region in inpainting. Secondly, the decoder is much heavier than the encoder, allowing for more detailed recovery while only slightly increasing the total inference cost. The training cost of our asymmetric VQGAN is cheap, and we only need to retrain a new asymmetric decoder while keeping the vanilla VQGAN encoder and StableDiffusion unchanged. Our asymmetric VQGAN can be widely used in StableDiffusion-based inpainting and local editing methods. Extensive experiments demonstrate that it can significantly improve the inpainting and editing performance, while maintaining the original text-to-image capability. The code is available at \url{https://github.com/buxiangzhiren/Asymmetric_VQGAN}.

Qinhong Yang, Dongdong Chen, Zhentao Tan et al.

This paper introduces a new large-scale image restoration dataset, called HQ-50K, which contains 50,000 high-quality images with rich texture details and semantic diversity. We analyze existing image restoration datasets from five different perspectives, including data scale, resolution, compression rates, texture details, and semantic coverage. However, we find that all of these datasets are deficient in some aspects. In contrast, HQ-50K considers all of these five aspects during the data curation process and meets all requirements. We also present a new Degradation-Aware Mixture of Expert (DAMoE) model, which enables a single model to handle multiple corruption types and unknown levels. Our extensive experiments demonstrate that HQ-50K consistently improves the performance on various image restoration tasks, such as super-resolution, denoising, dejpeg, and deraining. Furthermore, our proposed DAMoE, trained on our \dataset, outperforms existing state-of-the-art unified models designed for multiple restoration tasks and levels. The dataset and code are available at \url{https://github.com/littleYaang/HQ-50K}.

Yiting Cheng, Fangyun Wei, Jianmin Bao et al.

This work focuses on sign language retrieval-a recently proposed task for sign language understanding. Sign language retrieval consists of two sub-tasks: text-to-sign-video (T2V) retrieval and sign-video-to-text (V2T) retrieval. Different from traditional video-text retrieval, sign language videos, not only contain visual signals but also carry abundant semantic meanings by themselves due to the fact that sign languages are also natural languages. Considering this character, we formulate sign language retrieval as a cross-lingual retrieval problem as well as a video-text retrieval task. Concretely, we take into account the linguistic properties of both sign languages and natural languages, and simultaneously identify the fine-grained cross-lingual (i.e., sign-to-word) mappings while contrasting the texts and the sign videos in a joint embedding space. This process is termed as cross-lingual contrastive learning. Another challenge is raised by the data scarcity issue-sign language datasets are orders of magnitude smaller in scale than that of speech recognition. We alleviate this issue by adopting a domain-agnostic sign encoder pre-trained on large-scale sign videos into the target domain via pseudo-labeling. Our framework, termed as domain-aware sign language retrieval via Cross-lingual Contrastive learning or CiCo for short, outperforms the pioneering method by large margins on various datasets, e.g., +22.4 T2V and +28.0 V2T R@1 improvements on How2Sign dataset, and +13.7 T2V and +17.1 V2T R@1 improvements on PHOENIX-2014T dataset. Code and models are available at: https://github.com/FangyunWei/SLRT.

Weilun Wang, Jianmin Bao, Wengang Zhou et al.

We present SinDiffusion, leveraging denoising diffusion models to capture internal distribution of patches from a single natural image. SinDiffusion significantly improves the quality and diversity of generated samples compared with existing GAN-based approaches. It is based on two core designs. First, SinDiffusion is trained with a single model at a single scale instead of multiple models with progressive growing of scales which serves as the default setting in prior work. This avoids the accumulation of errors, which cause characteristic artifacts in generated results. Second, we identify that a patch-level receptive field of the diffusion network is crucial and effective for capturing the image's patch statistics, therefore we redesign the network structure of the diffusion model. Coupling these two designs enables us to generate photorealistic and diverse images from a single image. Furthermore, SinDiffusion can be applied to various applications, i.e., text-guided image generation, and image outpainting, due to the inherent capability of diffusion models. Extensive experiments on a wide range of images demonstrate the superiority of our proposed method for modeling the patch distribution.

Zigang Geng, Binxin Yang, Tiankai Hang et al.

We present InstructDiffusion, a unifying and generic framework for aligning computer vision tasks with human instructions. Unlike existing approaches that integrate prior knowledge and pre-define the output space (e.g., categories and coordinates) for each vision task, we cast diverse vision tasks into a human-intuitive image-manipulating process whose output space is a flexible and interactive pixel space. Concretely, the model is built upon the diffusion process and is trained to predict pixels according to user instructions, such as encircling the man's left shoulder in red or applying a blue mask to the left car. InstructDiffusion could handle a variety of vision tasks, including understanding tasks (such as segmentation and keypoint detection) and generative tasks (such as editing and enhancement). It even exhibits the ability to handle unseen tasks and outperforms prior methods on novel datasets. This represents a significant step towards a generalist modeling interface for vision tasks, advancing artificial general intelligence in the field of computer vision.

Zhicong Tang, Shuyang Gu, Jianmin Bao et al.

Vector quantized diffusion (VQ-Diffusion) is a powerful generative model for text-to-image synthesis, but sometimes can still generate low-quality samples or weakly correlated images with text input. We find these issues are mainly due to the flawed sampling strategy. In this paper, we propose two important techniques to further improve the sample quality of VQ-Diffusion. 1) We explore classifier-free guidance sampling for discrete denoising diffusion model and propose a more general and effective implementation of classifier-free guidance. 2) We present a high-quality inference strategy to alleviate the joint distribution issue in VQ-Diffusion. Finally, we conduct experiments on various datasets to validate their effectiveness and show that the improved VQ-Diffusion suppresses the vanilla version by large margins. We achieve an 8.44 FID score on MSCOCO, surpassing VQ-Diffusion by 5.42 FID score. When trained on ImageNet, we dramatically improve the FID score from 11.89 to 4.83, demonstrating the superiority of our proposed techniques.

Yiwei Ding, Wenjin Deng, Yinglin Zheng et al.

In this paper, we present the Intra- and Inter-Human Relation Networks (I^2R-Net) for Multi-Person Pose Estimation. It involves two basic modules. First, the Intra-Human Relation Module operates on a single person and aims to capture Intra-Human dependencies. Second, the Inter-Human Relation Module considers the relation between multiple instances and focuses on capturing Inter-Human interactions. The Inter-Human Relation Module can be designed very lightweight by reducing the resolution of feature map, yet learn useful relation information to significantly boost the performance of the Intra-Human Relation Module. Even without bells and whistles, our method can compete or outperform current competition winners. We conduct extensive experiments on COCO, CrowdPose, and OCHuman datasets. The results demonstrate that the proposed model surpasses all the state-of-the-art methods. Concretely, the proposed method achieves 77.4% AP on CrowPose dataset and 67.8% AP on OCHuman dataset respectively, outperforming existing methods by a large margin. Additionally, the ablation study and visualization analysis also prove the effectiveness of our model.

Yixuan Wang, Wengang Zhou, Jianmin Bao et al.

In this work, we are dedicated to text-guided image generation and propose a novel framework, i.e., CLIP2GAN, by leveraging CLIP model and StyleGAN. The key idea of our CLIP2GAN is to bridge the output feature embedding space of CLIP and the input latent space of StyleGAN, which is realized by introducing a mapping network. In the training stage, we encode an image with CLIP and map the output feature to a latent code, which is further used to reconstruct the image. In this way, the mapping network is optimized in a self-supervised learning way. In the inference stage, since CLIP can embed both image and text into a shared feature embedding space, we replace CLIP image encoder in the training architecture with CLIP text encoder, while keeping the following mapping network as well as StyleGAN model. As a result, we can flexibly input a text description to generate an image. Moreover, by simply adding mapped text features of an attribute to a mapped CLIP image feature, we can effectively edit the attribute to the image. Extensive experiments demonstrate the superior performance of our proposed CLIP2GAN compared to previous methods.

Wenming Weng, Ruoyu Feng, Yanhui Wang et al.

We present ART$\boldsymbol{\cdot}$V, an efficient framework for auto-regressive video generation with diffusion models. Unlike existing methods that generate entire videos in one-shot, ART$\boldsymbol{\cdot}$V generates a single frame at a time, conditioned on the previous ones. The framework offers three distinct advantages. First, it only learns simple continual motions between adjacent frames, therefore avoiding modeling complex long-range motions that require huge training data. Second, it preserves the high-fidelity generation ability of the pre-trained image diffusion models by making only minimal network modifications. Third, it can generate arbitrarily long videos conditioned on a variety of prompts such as text, image or their combinations, making it highly versatile and flexible. To combat the common drifting issue in AR models, we propose masked diffusion model which implicitly learns which information can be drawn from reference images rather than network predictions, in order to reduce the risk of generating inconsistent appearances that cause drifting. Moreover, we further enhance generation coherence by conditioning it on the initial frame, which typically contains minimal noise. This is particularly useful for long video generation. When trained for only two weeks on four GPUs, ART$\boldsymbol{\cdot}$V already can generate videos with natural motions, rich details and a high level of aesthetic quality. Besides, it enables various appealing applications, e.g., composing a long video from multiple text prompts.

Hezhen Hu, Xiaoyi Dong, Jianmin Bao et al.

Pre-training is playing an increasingly important role in learning generic feature representation for Person Re-identification (ReID). We argue that a high-quality ReID representation should have three properties, namely, multi-level awareness, occlusion robustness, and cross-region invariance. To this end, we propose a simple yet effective pre-training framework, namely PersonMAE, which involves two core designs into masked autoencoders to better serve the task of Person Re-ID. 1) PersonMAE generates two regions from the given image with RegionA as the input and \textit{RegionB} as the prediction target. RegionA is corrupted with block-wise masking to mimic common occlusion in ReID and its remaining visible parts are fed into the encoder. 2) Then PersonMAE aims to predict the whole RegionB at both pixel level and semantic feature level. It encourages its pre-trained feature representations with the three properties mentioned above. These properties make PersonMAE compatible with downstream Person ReID tasks, leading to state-of-the-art performance on four downstream ReID tasks, i.e., supervised (holistic and occluded setting), and unsupervised (UDA and USL setting). Notably, on the commonly adopted supervised setting, PersonMAE with ViT-B backbone achieves 79.8% and 69.5% mAP on the MSMT17 and OccDuke datasets, surpassing the previous state-of-the-art by a large margin of +8.0 mAP, and +5.3 mAP, respectively.

Marah Abdin, Jyoti Aneja, Hany Awadalla et al. · microsoft-research, stanford

We introduce phi-3-mini, a 3.8 billion parameter language model trained on 3.3 trillion tokens, whose overall performance, as measured by both academic benchmarks and internal testing, rivals that of models such as Mixtral 8x7B and GPT-3.5 (e.g., phi-3-mini achieves 69% on MMLU and 8.38 on MT-bench), despite being small enough to be deployed on a phone. Our training dataset is a scaled-up version of the one used for phi-2, composed of heavily filtered publicly available web data and synthetic data. The model is also further aligned for robustness, safety, and chat format. We also provide parameter-scaling results with a 7B, 14B models trained for 4.8T tokens, called phi-3-small, phi-3-medium, both significantly more capable than phi-3-mini (e.g., respectively 75%, 78% on MMLU, and 8.7, 8.9 on MT-bench). To enhance multilingual, multimodal, and long-context capabilities, we introduce three models in the phi-3.5 series: phi-3.5-mini, phi-3.5-MoE, and phi-3.5-Vision. The phi-3.5-MoE, a 16 x 3.8B MoE model with 6.6 billion active parameters, achieves superior performance in language reasoning, math, and code tasks compared to other open-source models of similar scale, such as Llama 3.1 and the Mixtral series, and on par with Gemini-1.5-Flash and GPT-4o-mini. Meanwhile, phi-3.5-Vision, a 4.2 billion parameter model derived from phi-3.5-mini, excels in reasoning tasks and is adept at handling both single-image and text prompts, as well as multi-image and text prompts.

Ruoyu Feng, Wenming Weng, Yanhui Wang et al.

In this paper, we present CCEdit, a versatile generative video editing framework based on diffusion models. Our approach employs a novel trident network structure that separates structure and appearance control, ensuring precise and creative editing capabilities. Utilizing the foundational ControlNet architecture, we maintain the structural integrity of the video during editing. The incorporation of an additional appearance branch enables users to exert fine-grained control over the edited key frame. These two side branches seamlessly integrate into the main branch, which is constructed upon existing text-to-image (T2I) generation models, through learnable temporal layers. The versatility of our framework is demonstrated through a diverse range of choices in both structure representations and personalized T2I models, as well as the option to provide the edited key frame. To facilitate comprehensive evaluation, we introduce the BalanceCC benchmark dataset, comprising 100 videos and 4 target prompts for each video. Our extensive user studies compare CCEdit with eight state-of-the-art video editing methods. The outcomes demonstrate CCEdit's substantial superiority over all other methods.

Pan Zhang, Jianmin Bao, Ting Zhang et al.

Recent image-to-image translation works have been transferred from supervised to unsupervised settings due to the expensive cost of capturing or labeling large amounts of paired data. However, current unsupervised methods using the cycle-consistency constraint may not find the desired mapping, especially for difficult translation tasks. On the other hand, a small number of paired data are usually accessible. We therefore introduce a general framework for semi-supervised image translation. Unlike previous works, our main idea is to learn the translation over the latent feature space instead of the image space. Thanks to the low dimensional feature space, it is easier to find the desired mapping function, resulting in improved quality of translation results as well as the stability of the translation model. Empirically we show that using feature translation generates better results, even using a few bits of paired data. Experimental comparisons with state-of-the-art approaches demonstrate the effectiveness of the proposed framework on a variety of challenging image-to-image translation tasks

Yanhui Wang, Jianmin Bao, Wenming Weng et al.

We present MicroCinema, a straightforward yet effective framework for high-quality and coherent text-to-video generation. Unlike existing approaches that align text prompts with video directly, MicroCinema introduces a Divide-and-Conquer strategy which divides the text-to-video into a two-stage process: text-to-image generation and image\&text-to-video generation. This strategy offers two significant advantages. a) It allows us to take full advantage of the recent advances in text-to-image models, such as Stable Diffusion, Midjourney, and DALLE, to generate photorealistic and highly detailed images. b) Leveraging the generated image, the model can allocate less focus to fine-grained appearance details, prioritizing the efficient learning of motion dynamics. To implement this strategy effectively, we introduce two core designs. First, we propose the Appearance Injection Network, enhancing the preservation of the appearance of the given image. Second, we introduce the Appearance Noise Prior, a novel mechanism aimed at maintaining the capabilities of pre-trained 2D diffusion models. These design elements empower MicroCinema to generate high-quality videos with precise motion, guided by the provided text prompts. Extensive experiments demonstrate the superiority of the proposed framework. Concretely, MicroCinema achieves SOTA zero-shot FVD of 342.86 on UCF-101 and 377.40 on MSR-VTT. See https://wangyanhui666.github.io/MicroCinema.github.io/ for video samples.

Mengchen Liu, Qixiu Li, Dongdong Chen et al.

With the release of GPT-4V(O), its use in generating pseudo labels for multi-modality tasks has gained significant popularity. However, it is still a secret how to build such advanced models from its base large language models (LLMs). This work explores the potential of using LLMs alone for data generation and develop competitive multi-modality models focusing on chart understanding. We construct a large-scale chart dataset, SynChart, which contains approximately 4 million diverse chart images with over 75 million dense annotations, including data tables, code, descriptions, and question-answer sets. We trained a 4.2B chart-expert model using this dataset and achieve near-GPT-4O performance on the ChartQA task, surpassing GPT-4V.

Dong Chen, Fangyun Wei, Ziyu Wan et al.

We introduce Lens, a 3.8B-parameter T2I model that achieves performance competitive with, and in several cases surpassing, state-of-the-art models with more than 6B parameters across various benchmarks, while requiring significantly less training compute. For example, Lens requires only about 19.3% of the training compute used by Z-Image. The training efficiency of Lens stems from two key strategies beyond its compact model size. First, we maximize data information density per training batch by (i) training on Lens-800M, a dataset of 800M densely captioned image-text pairs whose captions are generated by GPT-4.1 and contain approximately 109 words on average, providing richer semantic supervision than conventional short captions, and (ii) constructing each batch from images with multiple resolutions and diverse aspect ratios, thereby enlarging the effective visual coverage of each optimization step. Second, we improve convergence speed through careful architectural choices, including adopting a semantic VAE that provides better latent representations and employing a strong language encoder that accelerates optimization while enabling multilingual generalization from English-only training data. After pre-training, we apply RL with taxonomy-driven prompts (Lens-RL-8K) and structured reward rubrics to suppress artifacts and improve visual quality, a reasoner module with training-free system prompt search to better align user requests with the model, and distillation-based acceleration for 4-step inference. Through efficient training and systematic optimization, Lens generalizes to arbitrary aspect ratios from 1:2 to 2:1 and resolutions up to 1440^2, and supports prompts in several commonly used languages. Thanks to its compact size, Lens generates a 1024^2 image in 3.15 seconds on a single NVIDIA H100 GPU, while its distilled turbo version performs 4-step generation in 0.84 seconds.

Zhicong Tang, Jianmin Bao, Dong Chen et al.

This paper presents Model-guidance (MG), a novel objective for training diffusion model that addresses and removes of the commonly used Classifier-free guidance (CFG). Our innovative approach transcends the standard modeling of solely data distribution to incorporating the posterior probability of conditions. The proposed technique originates from the idea of CFG and is easy yet effective, making it a plug-and-play module for existing models. Our method significantly accelerates the training process, doubles the inference speed, and achieve exceptional quality that parallel and even surpass concurrent diffusion models with CFG. Extensive experiments demonstrate the effectiveness, efficiency, scalability on different models and datasets. Finally, we establish state-of-the-art performance on ImageNet 256 benchmarks with an FID of 1.34. Our code is available at https://github.com/tzco/Diffusion-wo-CFG.

Miaosen Zhang, Qi Dai, Yifan Yang et al.

LMMs have shown impressive visual understanding capabilities, with the potential to be applied in agents, which demand strong reasoning and planning abilities. Nevertheless, existing benchmarks mostly assess their reasoning abilities in language part, where the chain-of-thought is entirely composed of text.We consider the scenario where visual signals are continuously updated and required along the decision making process. Such vision-in-the-chain reasoning paradigm is more aligned with the needs of multimodal agents, while being rarely evaluated. In this paper, we introduce MageBench, a reasoning capability oriented multimodal agent benchmark that, while having light-weight environments, poses significant reasoning challenges and holds substantial practical value. This benchmark currently includes three types of environments: WebUI, Sokoban, and Football, comprising a total of 483 different scenarios. It thoroughly validates the agent's knowledge and engineering capabilities, visual intelligence, and interaction skills. The results show that only a few product-level models are better than random acting, and all of them are far inferior to human-level. More specifically, we found current models severely lack the ability to modify their planning based on visual feedback, as well as visual imagination, interleaved image-text long context handling, and other abilities. We hope that our work will provide optimization directions for LMM from the perspective of being an agent. We release our code and data at https://github.com/microsoft/MageBench.

Rui Tian, Qi Dai, Jianmin Bao et al.

Commercial video generation models have exhibited realistic, high-fidelity results but are still restricted to limited access. One crucial obstacle for large-scale applications is the expensive training and inference cost. In this paper, we argue that videos contain significantly more redundant information than images, allowing them to be encoded with very few motion latents. Towards this goal, we design an image-conditioned VAE that projects videos into extremely compressed latent space and decode them based on content images. This magic Reducio charm enables 64x reduction of latents compared to a common 2D VAE, without sacrificing the quality. Building upon Reducio-VAE, we can train diffusion models for high-resolution video generation efficiently. Specifically, we adopt a two-stage generation paradigm, first generating a condition image via text-to-image generation, followed by text-image-to-video generation with the proposed Reducio-DiT. Extensive experiments show that our model achieves strong performance in evaluation. More importantly, our method significantly boosts the training and inference efficiency of video LDMs. Reducio-DiT is trained in just 3.2K A100 GPU hours in total and can generate a 16-frame 1024$\times$1024 video clip within 15.5 seconds on a single A100 GPU. Code released at https://github.com/microsoft/Reducio-VAE .

Daichi Zhang, Tong Zhang, Jianmin Bao et al.

With the rapid development of generative models, detecting generated fake images to prevent their malicious use has become a critical issue recently. Existing methods frame this challenge as a naive binary image classification task. However, such methods focus only on visual clues, yielding trained detectors susceptible to overfitting specific image patterns and incapable of generalizing to unseen models. In this paper, we address this issue from a multi-modal perspective and find that fake images cannot be properly aligned with corresponding captions compared to real images. Upon this observation, we propose a simple yet effective detector termed ITEM by leveraging the image-text misalignment in a joint visual-language space as discriminative clues. Specifically, we first measure the misalignment of the images and captions in pre-trained CLIP's space, and then tune a MLP head to perform the usual detection task. Furthermore, we propose a hierarchical misalignment scheme that first focuses on the whole image and then each semantic object described in the caption, which can explore both global and fine-grained local semantic misalignment as clues. Extensive experiments demonstrate the superiority of our method against other state-of-the-art competitors with impressive generalization and robustness on various recent generative models.

Qixiu Li, Yaobo Liang, Zeyu Wang et al.

The advancement of large Vision-Language-Action (VLA) models has significantly improved robotic manipulation in terms of language-guided task execution and generalization to unseen scenarios. While existing VLAs adapted from pretrained large Vision-Language-Models (VLM) have demonstrated promising generalizability, their task performance is still unsatisfactory as indicated by the low tasks success rates in different environments. In this paper, we present a new advanced VLA architecture derived from VLM. Unlike previous works that directly repurpose VLM for action prediction by simple action quantization, we propose a omponentized VLA architecture that has a specialized action module conditioned on VLM output. We systematically study the design of the action module and demonstrates the strong performance enhancement with diffusion action transformers for action sequence modeling, as well as their favorable scaling behaviors. We also conduct comprehensive experiments and ablation studies to evaluate the efficacy of our models with varied designs. The evaluation on 5 robot embodiments in simulation and real work shows that our model not only significantly surpasses existing VLAs in task performance and but also exhibits remarkable adaptation to new robots and generalization to unseen objects and backgrounds. It exceeds the average success rates of OpenVLA which has similar model size (7B) with ours by over 35% in simulated evaluation and 55% in real robot experiments. It also outperforms the large RT-2-X model (55B) by 18% absolute success rates in simulation. Code and models can be found on our project page (https://cogact.github.io/).

Shihao Zhao, Dongdong Chen, Yen-Chun Chen et al.

Text-to-Image diffusion models have made tremendous progress over the past two years, enabling the generation of highly realistic images based on open-domain text descriptions. However, despite their success, text descriptions often struggle to adequately convey detailed controls, even when composed of long and complex texts. Moreover, recent studies have also shown that these models face challenges in understanding such complex texts and generating the corresponding images. Therefore, there is a growing need to enable more control modes beyond text description. In this paper, we introduce Uni-ControlNet, a unified framework that allows for the simultaneous utilization of different local controls (e.g., edge maps, depth map, segmentation masks) and global controls (e.g., CLIP image embeddings) in a flexible and composable manner within one single model. Unlike existing methods, Uni-ControlNet only requires the fine-tuning of two additional adapters upon frozen pre-trained text-to-image diffusion models, eliminating the huge cost of training from scratch. Moreover, thanks to some dedicated adapter designs, Uni-ControlNet only necessitates a constant number (i.e., 2) of adapters, regardless of the number of local or global controls used. This not only reduces the fine-tuning costs and model size, making it more suitable for real-world deployment, but also facilitate composability of different conditions. Through both quantitative and qualitative comparisons, Uni-ControlNet demonstrates its superiority over existing methods in terms of controllability, generation quality and composability. Code is available at \url{https://github.com/ShihaoZhaoZSH/Uni-ControlNet}.

Dengpan Fu, Dongdong Chen, Hao Yang et al.

This paper aims to address the problem of pre-training for person re-identification (Re-ID) with noisy labels. To setup the pre-training task, we apply a simple online multi-object tracking system on raw videos of an existing unlabeled Re-ID dataset "LUPerson" nd build the Noisy Labeled variant called "LUPerson-NL". Since theses ID labels automatically derived from tracklets inevitably contain noises, we develop a large-scale Pre-training framework utilizing Noisy Labels (PNL), which consists of three learning modules: supervised Re-ID learning, prototype-based contrastive learning, and label-guided contrastive learning. In principle, joint learning of these three modules not only clusters similar examples to one prototype, but also rectifies noisy labels based on the prototype assignment. We demonstrate that learning directly from raw videos is a promising alternative for pre-training, which utilizes spatial and temporal correlations as weak supervision. This simple pre-training task provides a scalable way to learn SOTA Re-ID representations from scratch on "LUPerson-NL" without bells and whistles. For example, by applying on the same supervised Re-ID method MGN, our pre-trained model improves the mAP over the unsupervised pre-training counterpart by 5.7%, 2.2%, 2.3% on CUHK03, DukeMTMC, and MSMT17 respectively. Under the small-scale or few-shot setting, the performance gain is even more significant, suggesting a better transferability of the learned representation. Code is available at https://github.com/DengpanFu/LUPerson-NL

Bowen Zhang, Shuyang Gu, Bo Zhang et al.

Despite the tantalizing success in a broad of vision tasks, transformers have not yet demonstrated on-par ability as ConvNets in high-resolution image generative modeling. In this paper, we seek to explore using pure transformers to build a generative adversarial network for high-resolution image synthesis. To this end, we believe that local attention is crucial to strike the balance between computational efficiency and modeling capacity. Hence, the proposed generator adopts Swin transformer in a style-based architecture. To achieve a larger receptive field, we propose double attention which simultaneously leverages the context of the local and the shifted windows, leading to improved generation quality. Moreover, we show that offering the knowledge of the absolute position that has been lost in window-based transformers greatly benefits the generation quality. The proposed StyleSwin is scalable to high resolutions, with both the coarse geometry and fine structures benefit from the strong expressivity of transformers. However, blocking artifacts occur during high-resolution synthesis because performing the local attention in a block-wise manner may break the spatial coherency. To solve this, we empirically investigate various solutions, among which we find that employing a wavelet discriminator to examine the spectral discrepancy effectively suppresses the artifacts. Extensive experiments show the superiority over prior transformer-based GANs, especially on high resolutions, e.g., 1024x1024. The StyleSwin, without complex training strategies, excels over StyleGAN on CelebA-HQ 1024, and achieves on-par performance on FFHQ-1024, proving the promise of using transformers for high-resolution image generation. The code and models will be available at https://github.com/microsoft/StyleSwin.

Zhenda Xie, Zheng Zhang, Yue Cao et al.

This paper presents SimMIM, a simple framework for masked image modeling. We simplify recently proposed related approaches without special designs such as block-wise masking and tokenization via discrete VAE or clustering. To study what let the masked image modeling task learn good representations, we systematically study the major components in our framework, and find that simple designs of each component have revealed very strong representation learning performance: 1) random masking of the input image with a moderately large masked patch size (e.g., 32) makes a strong pre-text task; 2) predicting raw pixels of RGB values by direct regression performs no worse than the patch classification approaches with complex designs; 3) the prediction head can be as light as a linear layer, with no worse performance than heavier ones. Using ViT-B, our approach achieves 83.8% top-1 fine-tuning accuracy on ImageNet-1K by pre-training also on this dataset, surpassing previous best approach by +0.6%. When applied on a larger model of about 650 million parameters, SwinV2-H, it achieves 87.1% top-1 accuracy on ImageNet-1K using only ImageNet-1K data. We also leverage this approach to facilitate the training of a 3B model (SwinV2-G), that by $40\times$ less data than that in previous practice, we achieve the state-of-the-art on four representative vision benchmarks. The code and models will be publicly available at https://github.com/microsoft/SimMIM.

Yiting Cheng, Fangyun Wei, Jianmin Bao et al.

Domain adaptation for semantic segmentation enables to alleviate the need for large-scale pixel-wise annotations. Recently, self-supervised learning (SSL) with a combination of image-to-image translation shows great effectiveness in adaptive segmentation. The most common practice is to perform SSL along with image translation to well align a single domain (the source or target). However, in this single-domain paradigm, unavoidable visual inconsistency raised by image translation may affect subsequent learning. In this paper, based on the observation that domain adaptation frameworks performed in the source and target domain are almost complementary in terms of image translation and SSL, we propose a novel dual path learning (DPL) framework to alleviate visual inconsistency. Concretely, DPL contains two complementary and interactive single-domain adaptation pipelines aligned in source and target domain respectively. The inference of DPL is extremely simple, only one segmentation model in the target domain is employed. Novel technologies such as dual path image translation and dual path adaptive segmentation are proposed to make two paths promote each other in an interactive manner. Experiments on GTA5$\rightarrow$Cityscapes and SYNTHIA$\rightarrow$Cityscapes scenarios demonstrate the superiority of our DPL model over the state-of-the-art methods. The code and models are available at: \url{https://github.com/royee182/DPL}

Xiaoyi Dong, Jianmin Bao, Dongdong Chen et al.

We present CSWin Transformer, an efficient and effective Transformer-based backbone for general-purpose vision tasks. A challenging issue in Transformer design is that global self-attention is very expensive to compute whereas local self-attention often limits the field of interactions of each token. To address this issue, we develop the Cross-Shaped Window self-attention mechanism for computing self-attention in the horizontal and vertical stripes in parallel that form a cross-shaped window, with each stripe obtained by splitting the input feature into stripes of equal width. We provide a mathematical analysis of the effect of the stripe width and vary the stripe width for different layers of the Transformer network which achieves strong modeling capability while limiting the computation cost. We also introduce Locally-enhanced Positional Encoding (LePE), which handles the local positional information better than existing encoding schemes. LePE naturally supports arbitrary input resolutions, and is thus especially effective and friendly for downstream tasks. Incorporated with these designs and a hierarchical structure, CSWin Transformer demonstrates competitive performance on common vision tasks. Specifically, it achieves 85.4\% Top-1 accuracy on ImageNet-1K without any extra training data or label, 53.9 box AP and 46.4 mask AP on the COCO detection task, and 52.2 mIOU on the ADE20K semantic segmentation task, surpassing previous state-of-the-art Swin Transformer backbone by +1.2, +2.0, +1.4, and +2.0 respectively under the similar FLOPs setting. By further pretraining on the larger dataset ImageNet-21K, we achieve 87.5% Top-1 accuracy on ImageNet-1K and high segmentation performance on ADE20K with 55.7 mIoU. The code and models are available at https://github.com/microsoft/CSWin-Transformer.

Zhendong Wang, Xiaodong Cun, Jianmin Bao et al.

In this paper, we present Uformer, an effective and efficient Transformer-based architecture for image restoration, in which we build a hierarchical encoder-decoder network using the Transformer block. In Uformer, there are two core designs. First, we introduce a novel locally-enhanced window (LeWin) Transformer block, which performs nonoverlapping window-based self-attention instead of global self-attention. It significantly reduces the computational complexity on high resolution feature map while capturing local context. Second, we propose a learnable multi-scale restoration modulator in the form of a multi-scale spatial bias to adjust features in multiple layers of the Uformer decoder. Our modulator demonstrates superior capability for restoring details for various image restoration tasks while introducing marginal extra parameters and computational cost. Powered by these two designs, Uformer enjoys a high capability for capturing both local and global dependencies for image restoration. To evaluate our approach, extensive experiments are conducted on several image restoration tasks, including image denoising, motion deblurring, defocus deblurring and deraining. Without bells and whistles, our Uformer achieves superior or comparable performance compared with the state-of-the-art algorithms. The code and models are available at https://github.com/ZhendongWang6/Uformer.

Xiaoyi Dong, Dongdong Chen, Jianmin Bao et al.

Modern deep neural networks(DNNs) are vulnerable to adversarial samples. Sparse adversarial samples are a special branch of adversarial samples that can fool the target model by only perturbing a few pixels. The existence of the sparse adversarial attack points out that DNNs are much more vulnerable than people believed, which is also a new aspect for analyzing DNNs. However, current sparse adversarial attack methods still have some shortcomings on both sparsity and invisibility. In this paper, we propose a novel two-stage distortion-aware greedy-based method dubbed as "GreedyFool". Specifically, it first selects the most effective candidate positions to modify by considering both the gradient(for adversary) and the distortion map(for invisibility), then drops some less important points in the reduce stage. Experiments demonstrate that compared with the start-of-the-art method, we only need to modify $3\times$ fewer pixels under the same sparse perturbation setting. For target attack, the success rate of our method is 9.96\% higher than the start-of-the-art method under the same pixel budget. Code can be found at https://github.com/LightDXY/GreedyFool.

Zhicong Tang, Shuyang Gu, Chunyu Wang et al.

This paper introduces a pioneering 3D volumetric encoder designed for text-to-3D generation. To scale up the training data for the diffusion model, a lightweight network is developed to efficiently acquire feature volumes from multi-view images. The 3D volumes are then trained on a diffusion model for text-to-3D generation using a 3D U-Net. This research further addresses the challenges of inaccurate object captions and high-dimensional feature volumes. The proposed model, trained on the public Objaverse dataset, demonstrates promising outcomes in producing diverse and recognizable samples from text prompts. Notably, it empowers finer control over object part characteristics through textual cues, fostering model creativity by seamlessly combining multiple concepts within a single object. This research significantly contributes to the progress of 3D generation by introducing an efficient, flexible, and scalable representation methodology.

Yifan Pu, Yiming Zhao, Zhicong Tang et al.

Multi-layer image generation is a fundamental task that enables users to isolate, select, and edit specific image layers, thereby revolutionizing interactions with generative models. In this paper, we introduce the Anonymous Region Transformer (ART), which facilitates the direct generation of variable multi-layer transparent images based on a global text prompt and an anonymous region layout. Inspired by Schema theory suggests that knowledge is organized in frameworks (schemas) that enable people to interpret and learn from new information by linking it to prior knowledge.}, this anonymous region layout allows the generative model to autonomously determine which set of visual tokens should align with which text tokens, which is in contrast to the previously dominant semantic layout for the image generation task. In addition, the layer-wise region crop mechanism, which only selects the visual tokens belonging to each anonymous region, significantly reduces attention computation costs and enables the efficient generation of images with numerous distinct layers (e.g., 50+). When compared to the full attention approach, our method is over 12 times faster and exhibits fewer layer conflicts. Furthermore, we propose a high-quality multi-layer transparent image autoencoder that supports the direct encoding and decoding of the transparency of variable multi-layer images in a joint manner. By enabling precise control and scalable layer generation, ART establishes a new paradigm for interactive content creation.

Zhendong Wang, Jianmin Bao, Shuyang Gu et al.

In this paper, we present DesignDiffusion, a simple yet effective framework for the novel task of synthesizing design images from textual descriptions. A primary challenge lies in generating accurate and style-consistent textual and visual content. Existing works in a related task of visual text generation often focus on generating text within given specific regions, which limits the creativity of generation models, resulting in style or color inconsistencies between textual and visual elements if applied to design image generation. To address this issue, we propose an end-to-end, one-stage diffusion-based framework that avoids intricate components like position and layout modeling. Specifically, the proposed framework directly synthesizes textual and visual design elements from user prompts. It utilizes a distinctive character embedding derived from the visual text to enhance the input prompt, along with a character localization loss for enhanced supervision during text generation. Furthermore, we employ a self-play Direct Preference Optimization fine-tuning strategy to improve the quality and accuracy of the synthesized visual text. Extensive experiments demonstrate that DesignDiffusion achieves state-of-the-art performance in design image generation.

Dianmo Sheng, Dongdong Chen, Zhentao Tan et al.

The rapid advancement of large language models (LLMs) has accelerated the emergence of in-context learning (ICL) as a cutting-edge approach in the natural language processing domain. Recently, ICL has been employed in visual understanding tasks, such as semantic segmentation and image captioning, yielding promising results. However, existing visual ICL framework can not enable producing content across multiple modalities, which limits their potential usage scenarios. To address this issue, we present a new ICL framework for visual understanding with multi-modal output enabled. First, we quantize and embed both text and visual prompt into a unified representational space, structured as interleaved in-context sequences. Then a decoder-only sparse transformer architecture is employed to perform generative modeling on them, facilitating in-context learning. Thanks to this design, the model is capable of handling in-context vision understanding tasks with multimodal output in a unified pipeline.Experimental results demonstrate that our model achieves competitive performance compared with specialized models and previous ICL baselines. Overall, our research takes a further step toward unified multimodal in-context learning.

Yu Zhang, Yunqi Li, Yifan Yang et al.

Although chain-of-thought reasoning and reinforcement learning (RL) have driven breakthroughs in NLP, their integration into generative vision models remains underexplored. We introduce ReasonGen-R1, a two-stage framework that first imbues an autoregressive image generator with explicit text-based "thinking" skills via supervised fine-tuning on a newly generated reasoning dataset of written rationales, and then refines its outputs using Group Relative Policy Optimization. To enable the model to reason through text before generating images, We automatically generate and release a corpus of model crafted rationales paired with visual prompts, enabling controlled planning of object layouts, styles, and scene compositions. Our GRPO algorithm uses reward signals from a pretrained vision language model to assess overall visual quality, optimizing the policy in each update. Evaluations on GenEval, DPG, and the T2I benchmark demonstrate that ReasonGen-R1 consistently outperforms strong baselines and prior state-of-the-art models. More: aka.ms/reasongen.

Longtao Jiang, Zhendong Wang, Jianmin Bao et al.

Object removal has so far been dominated by the mask-and-inpaint paradigm, where the masked region is excluded from the input, leaving models relying on unmasked areas to inpaint the missing region. However, this approach lacks contextual information for the masked area, often resulting in unstable performance. In this work, we introduce SmartEraser, built with a new removing paradigm called Masked-Region Guidance. This paradigm retains the masked region in the input, using it as guidance for the removal process. It offers several distinct advantages: (a) it guides the model to accurately identify the object to be removed, preventing its regeneration in the output; (b) since the user mask often extends beyond the object itself, it aids in preserving the surrounding context in the final result. Leveraging this new paradigm, we present Syn4Removal, a large-scale object removal dataset, where instance segmentation data is used to copy and paste objects onto images as removal targets, with the original images serving as ground truths. Experimental results demonstrate that SmartEraser significantly outperforms existing methods, achieving superior performance in object removal, especially in complex scenes with intricate compositions.

Tiankai Hang, Jianmin Bao, Fangyun Wei et al.

Autoregressive models have demonstrated remarkable success in sequential data generation, particularly in NLP, but their extension to continuous-domain image generation presents significant challenges. Recent work, the masked autoregressive model (MAR), bypasses quantization by modeling per-token distributions in continuous spaces using a diffusion head but suffers from slow inference due to the high computational cost of the iterative denoising process. To address this, we propose the Fast AutoRegressive model (FAR), a novel framework that replaces MAR's diffusion head with a lightweight shortcut head, enabling efficient few-step sampling while preserving autoregressive principles. Additionally, FAR seamlessly integrates with causal Transformers, extending them from discrete to continuous token generation without requiring architectural modifications. Experiments demonstrate that FAR achieves $2.3\times$ faster inference than MAR while maintaining competitive FID and IS scores. This work establishes the first efficient autoregressive paradigm for high-fidelity continuous-space image generation, bridging the critical gap between quality and scalability in visual autoregressive modeling.

Wangzheng Shi, Yinglin Zheng, Yuxin Lin et al.

Hair transfer is increasingly valuable across domains such as social media, gaming, advertising, and entertainment. While significant progress has been made in single-image hair transfer, video-based hair transfer remains challenging due to the need for temporal consistency, spatial fidelity, and dynamic adaptability. In this work, we propose HairShifter, a novel "Anchor Frame + Animation" framework that unifies high-quality image hair transfer with smooth and coherent video animation. At its core, HairShifter integrates a Image Hair Transfer (IHT) module for precise per-frame transformation and a Multi-Scale Gated SPADE Decoder to ensure seamless spatial blending and temporal coherence. Our method maintains hairstyle fidelity across frames while preserving non-hair regions. Extensive experiments demonstrate that HairShifter achieves state-of-the-art performance in video hairstyle transfer, combining superior visual quality, temporal consistency, and scalability. The code will be publicly available. We believe this work will open new avenues for video-based hairstyle transfer and establish a robust baseline in this field.

Xinzhi Mu, Li Chen, Bohan Chen et al.

Recently, the application of modern diffusion-based text-to-image generation models for creating artistic fonts, traditionally the domain of professional designers, has garnered significant interest. Diverging from the majority of existing studies that concentrate on generating artistic typography, our research aims to tackle a novel and more demanding challenge: the generation of text effects for multilingual fonts. This task essentially requires generating coherent and consistent visual content within the confines of a font-shaped canvas, as opposed to a traditional rectangular canvas. To address this task, we introduce a novel shape-adaptive diffusion model capable of interpreting the given shape and strategically planning pixel distributions within the irregular canvas. To achieve this, we curate a high-quality shape-adaptive image-text dataset and incorporate the segmentation mask as a visual condition to steer the image generation process within the irregular-canvas. This approach enables the traditionally rectangle canvas-based diffusion model to produce the desired concepts in accordance with the provided geometric shapes. Second, to maintain consistency across multiple letters, we also present a training-free, shape-adaptive effect transfer method for transferring textures from a generated reference letter to others. The key insights are building a font effect noise prior and propagating the font effect information in a concatenated latent space. The efficacy of our FontStudio system is confirmed through user preference studies, which show a marked preference (78% win-rates on aesthetics) for our system even when compared to the latest unrivaled commercial product, Adobe Firefly.