Recently, medical image synthesis gains more and more popularity, along with the rapid development of generative models. Medical image synthesis aims to generate an unacquired image modality, often from other observed data modalities. Synthesized images can be used for clinical diagnostic assistance, data augmentation for model training and validation or image quality improving. In the meanwhile, the flow-based models are among the successful generative models for the ability of generating realistic and high-quality synthetic images. However, most flow-based models require to calculate flow ordinary different equation (ODE) evolution steps in transfer process, for which the performances are significantly limited by heavy computation time due to a large number of time iterations. In this paper, we propose a novel flow-based model, namely Discrete Process Matching (DPM) to accomplish the bi-modality image transfer tasks. Different to other flow matching based models, we propose to utilize both forward and backward ODE flow and enhance the consistency on the intermediate images of few discrete time steps, resulting in a transfer process with much less iteration steps while maintaining high-quality generations for both modalities. Our experiments on three datasets of MRI T1/T2 and CT/MRI demonstrate that DPM outperforms other state-of-the-art flow-based methods for bi-modality image synthesis, achieving higher image quality with less computation time cost.
Recent studies have demonstrated the effectiveness of token-based methods for visual content generation. As a representative work, non-autoregressive Transformers (NATs) are able to synthesize images with decent quality in a small number of steps. However, NATs usually necessitate configuring a complicated generation policy comprising multiple manually-designed scheduling rules. These heuristic-driven rules are prone to sub-optimality and come with the requirements of expert knowledge and labor-intensive efforts. Moreover, their one-size-fits-all nature cannot flexibly adapt to the diverse characteristics of each individual sample. To address these issues, we propose AdaNAT, a learnable approach that automatically configures a suitable policy tailored for every sample to be generated. In specific, we formulate the determination of generation policies as a Markov decision process. Under this framework, a lightweight policy network for generation can be learned via reinforcement learning. Importantly, we demonstrate that simple reward designs such as FID or pre-trained reward models, may not reliably guarantee the desired quality or diversity of generated samples. Therefore, we propose an adversarial reward design to guide the training of policy networks effectively. Comprehensive experiments on four benchmark datasets, i.e., ImageNet-256 & 512, MS-COCO, and CC3M, validate the effectiveness of AdaNAT. Code and pre-trained models will be released at https://github.com/LeapLabTHU/AdaNAT.
We present Open-MAGVIT2, a family of auto-regressive image generation models ranging from 300M to 1.5B. The Open-MAGVIT2 project produces an open-source replication of Google's MAGVIT-v2 tokenizer, a tokenizer with a super-large codebook (i.e., $2^{18}$ codes), and achieves the state-of-the-art reconstruction performance (1.17 rFID) on ImageNet $256 \times 256$. Furthermore, we explore its application in plain auto-regressive models and validate scalability properties. To assist auto-regressive models in predicting with a super-large vocabulary, we factorize it into two sub-vocabulary of different sizes by asymmetric token factorization, and further introduce "next sub-token prediction" to enhance sub-token interaction for better generation quality. We release all models and codes to foster innovation and creativity in the field of auto-regressive visual generation.
Subject-driven text-to-image generation has witnessed remarkable advancements in its ability to learn and capture characteristics of a subject using only a limited number of images. However, existing methods commonly rely on high-quality images for training and may struggle to generate reasonable images when the input images are blemished by artifacts. This is primarily attributed to the inadequate capability of current techniques in distinguishing subject-related features from disruptive artifacts. In this paper, we introduce ArtiFade to tackle this issue and successfully generate high-quality artifact-free images from blemished datasets. Specifically, ArtiFade exploits fine-tuning of a pre-trained text-to-image model, aiming to remove artifacts. The elimination of artifacts is achieved by utilizing a specialized dataset that encompasses both unblemished images and their corresponding blemished counterparts during fine-tuning. ArtiFade also ensures the preservation of the original generative capabilities inherent within the diffusion model, thereby enhancing the overall performance of subject-driven methods in generating high-quality and artifact-free images. We further devise evaluation benchmarks tailored for this task. Through extensive qualitative and quantitative experiments, we demonstrate the generalizability of ArtiFade in effective artifact removal under both in-distribution and out-of-distribution scenarios.
Due to lack of fully publicly available text-to-video models, current video editing methods tend to build on pre-trained text-to-image generation models, however, they still face grand challenges in dealing with the local editing of video with temporal information. First, although existing methods attempt to focus on local area editing by a pre-defined mask, the preservation of the outside-area background is non-ideal due to the spatially entire generation of each frame. In addition, specially providing a mask by user is an additional costly undertaking, so an autonomous masking strategy integrated into the editing process is desirable. Last but not least, image-level pretrained model hasn't learned temporal information across frames of a video which is vital for expressing the motion and dynamics. In this paper, we propose to adapt a image-level blended latent diffusion model to perform local video editing tasks. Specifically, we leverage DDIM inversion to acquire the latents as background latents instead of the randomly noised ones to better preserve the background information of the input video. We further introduce an autonomous mask manufacture mechanism derived from cross-attention maps in diffusion steps. Finally, we enhance the temporal consistency across video frames by transforming the self-attention blocks of U-Net into temporal-spatial blocks. Through extensive experiments, our proposed approach demonstrates effectiveness in different real-world video editing tasks.
The advancement of Large Language Models (LLMs) for domain applications in fields such as materials science and engineering depends on the development of fine-tuning strategies that adapt models for specialized, technical capabilities. In this work, we explore the effects of Continued Pretraining (CPT), Supervised Fine-Tuning (SFT), and various preference-based optimization approaches, including Direct Preference Optimization (DPO) and Odds Ratio Preference Optimization (ORPO), on fine-tuned LLM performance. Our analysis shows how these strategies influence model outcomes and reveals that the merging of multiple fine-tuned models can lead to the emergence of capabilities that surpass the individual contributions of the parent models. We find that model merging leads to new functionalities that neither parent model could achieve alone, leading to improved performance in domain-specific assessments. Experiments with different model architectures are presented, including Llama 3.1 8B and Mistral 7B models, where similar behaviors are observed. Exploring whether the results hold also for much smaller models, we use a tiny LLM with 1.7 billion parameters and show that very small LLMs do not necessarily feature emergent capabilities under model merging, suggesting that model scaling may be a key component. In open-ended yet consistent chat conversations between a human and AI models, our assessment reveals detailed insights into how different model variants perform and show that the smallest model achieves a high intelligence score across key criteria including reasoning depth, creativity, clarity, and quantitative precision. Other experiments include the development of image generation prompts based on disparate biological material design concepts, to create new microstructures, architectural concepts, and urban design based on biological materials-inspired construction principles.
Stable diffusion has demonstrated strong image synthesis ability to given text descriptions, suggesting it to contain strong semantic clue for grouping objects. The researchers have explored employing stable diffusion for training-free segmentation. Most existing approaches refine cross-attention map by self-attention map once, demonstrating that self-attention map contains useful semantic information to improve segmentation. To fully utilize self-attention map, we present a deep experimental analysis on iteratively refining cross-attention map with self-attention map, and propose an effective iterative refinement framework for training-free segmentation, named iSeg. The proposed iSeg introduces an entropy-reduced self-attention module that utilizes a gradient descent scheme to reduce the entropy of self-attention map, thereby suppressing the weak responses corresponding to irrelevant global information. Leveraging the entropy-reduced self-attention module, our iSeg stably improves refined cross-attention map with iterative refinement. Further, we design a category-enhanced cross-attention module to generate accurate cross-attention map, providing a better initial input for iterative refinement. Extensive experiments across different datasets and diverse segmentation tasks reveal the merits of proposed contributions, leading to promising performance on diverse segmentation tasks. For unsupervised semantic segmentation on Cityscapes, our iSeg achieves an absolute gain of 3.8% in terms of mIoU compared to the best existing training-free approach in literature. Moreover, our proposed iSeg can support segmentation with different kinds of images and interactions. The project is available at https://linsun449.github.io/iSeg.
Text-to-image diffusion models have emerged as a powerful framework for high-quality image generation given textual prompts. Their success has driven the rapid development of production-grade diffusion models that consistently increase in size and already contain billions of parameters. As a result, state-of-the-art text-to-image models are becoming less accessible in practice, especially in resource-limited environments. Post-training quantization (PTQ) tackles this issue by compressing the pretrained model weights into lower-bit representations. Recent diffusion quantization techniques primarily rely on uniform scalar quantization, providing decent performance for the models compressed to 4 bits. This work demonstrates that more versatile vector quantization (VQ) may achieve higher compression rates for large-scale text-to-image diffusion models. Specifically, we tailor vector-based PTQ methods to recent billion-scale text-to-image models (SDXL and SDXL-Turbo), and show that the diffusion models of 2B+ parameters compressed to around 3 bits using VQ exhibit the similar image quality and textual alignment as previous 4-bit compression techniques.
Recent text-to-image models have achieved remarkable success in generating high-quality images. However, when tasked with multi-concept generation which creates images containing multiple characters or objects, existing methods often suffer from attribute confusion, resulting in severe text-image inconsistency. We found that attribute confusion occurs when a certain region of the latent features attend to multiple or incorrect prompt tokens. In this work, we propose novel Semantic Protection Diffusion (SPDiffusion) to protect the semantics of regions from the influence of irrelevant tokens, eliminating the confusion of non-corresponding attributes. In the SPDiffusion framework, we design a Semantic Protection Mask (SP-Mask) to represent the relevance of the regions and the tokens, and propose a Semantic Protection Cross-Attention (SP-Attn) to shield the influence of irrelevant tokens on specific regions in the generation process. To evaluate our method, we created a diverse multi-concept benchmark, and SPDiffusion achieves state-of-the-art results on this benchmark, proving its effectiveness. Our method can be combined with many other application methods or backbones, such as ControlNet, Story Diffusion, PhotoMaker and PixArt-alpha to enhance their multi-concept capabilities, demonstrating strong compatibility and scalability.
The evaluation of image generators remains a challenge due to the limitations of traditional metrics in providing nuanced insights into specific image regions. This is a critical problem as not all regions of an image may be learned with similar ease. In this work, we propose a novel approach to disentangle the cosine similarity of mean embeddings into the product of cosine similarities for individual pixel clusters via central kernel alignment. Consequently, we can quantify the contribution of the cluster-wise performance to the overall image generation performance. We demonstrate how this enhances the explainability and the likelihood of identifying pixel regions of model misbehavior across various real-world use cases.
The potential for higher-resolution image generation using pretrained diffusion models is immense, yet these models often struggle with issues of object repetition and structural artifacts especially when scaling to 4K resolution and higher. We figure out that the problem is caused by that, a single prompt for the generation of multiple scales provides insufficient efficacy. In response, we propose HiPrompt, a new tuning-free solution that tackles the above problems by introducing hierarchical prompts. The hierarchical prompts offer both global and local guidance. Specifically, the global guidance comes from the user input that describes the overall content, while the local guidance utilizes patch-wise descriptions from MLLMs to elaborately guide the regional structure and texture generation. Furthermore, during the inverse denoising process, the generated noise is decomposed into low- and high-frequency spatial components. These components are conditioned on multiple prompt levels, including detailed patch-wise descriptions and broader image-level prompts, facilitating prompt-guided denoising under hierarchical semantic guidance. It further allows the generation to focus more on local spatial regions and ensures the generated images maintain coherent local and global semantics, structures, and textures with high definition. Extensive experiments demonstrate that HiPrompt outperforms state-of-the-art works in higher-resolution image generation, significantly reducing object repetition and enhancing structural quality.
The emergence of diffusion models has enabled the generation of diverse high-quality images solely from text, prompting subsequent efforts to enhance the controllability of these models. Despite the improvement in controllability, pose control remains limited to specific objects (e.g., humans) or poses (e.g., frontal view) due to the fact that pose is generally controlled via camera parameters (e.g., rotation angle) or keypoints (e.g., eyes, nose). Specifically, camera parameters-conditional pose control models generate unrealistic images depending on the object, owing to the small size of 3D datasets for training. Also, keypoint-based approaches encounter challenges in acquiring reliable keypoints for various objects (e.g., church) or poses (e.g., back view). To address these limitations, we propose depth-based pose control, as depth maps are easily obtainable from a single depth estimation model regardless of objects and poses, unlike camera parameters and keypoints. However, depth-based pose control confronts issues of shape dependency, as depth maps influence not only the pose but also the shape of the generated images. To tackle this issue, we propose Skip-and-Play (SnP), designed via analysis of the impact of three components of depth-conditional ControlNet on the pose and the shape of the generated images. To be specific, based on the analysis, we selectively skip parts of the components to mitigate shape dependency on the depth map while preserving the pose. Through various experiments, we demonstrate the superiority of SnP over baselines and showcase the ability of SnP to generate images of diverse objects and poses. Remarkably, SnP exhibits the ability to generate images even when the objects in the condition (e.g., a horse) and the prompt (e.g., a hedgehog) differ from each other.
Despite the burst of innovative methods for controlling the diffusion process, effectively controlling image styles in text-to-image generation remains a challenging task. Many adapter-based methods impose image representation conditions on the denoising process to accomplish image control. However these conditions are not aligned with the word embedding space, leading to interference between image and text control conditions and the potential loss of semantic information from the text prompt. Addressing this issue involves two key challenges. Firstly, how to inject the style representation without compromising the effectiveness of text representation in control. Secondly, how to obtain the accurate style representation from a single reference image. To tackle these challenges, we introduce StyleTokenizer, a zero-shot style control image generation method that aligns style representation with text representation using a style tokenizer. This alignment effectively minimizes the impact on the effectiveness of text prompts. Furthermore, we collect a well-labeled style dataset named Style30k to train a style feature extractor capable of accurately representing style while excluding other content information. Experimental results demonstrate that our method fully grasps the style characteristics of the reference image, generating appealing images that are consistent with both the target image style and text prompt. The code and dataset are available at https://github.com/alipay/style-tokenizer.
We consider the problem of independently, in a disentangled fashion, controlling the outputs of text-to-image diffusion models with color and style attributes of a user-supplied reference image. We present the first training-free, test-time-only method to disentangle and condition text-to-image models on color and style attributes from reference image. To realize this, we propose two key innovations. Our first contribution is to transform the latent codes at inference time using feature transformations that make the covariance matrix of current generation follow that of the reference image, helping meaningfully transfer color. Next, we observe that there exists a natural disentanglement between color and style in the LAB image space, which we exploit to transform the self-attention feature maps of the image being generated with respect to those of the reference computed from its L channel. Both these operations happen purely at test time and can be done independently or merged. This results in a flexible method where color and style information can come from the same reference image or two different sources, and a new generation can seamlessly fuse them in either scenario.
Recently, diffusion models have emerged as a powerful class of generative models. Despite their success, there is still limited understanding of their semantic spaces. This makes it challenging to achieve precise and disentangled image generation without additional training, especially in an unsupervised way. In this work, we improve the understanding of their semantic spaces from intriguing observations: among a certain range of noise levels, (1) the learned posterior mean predictor (PMP) in the diffusion model is locally linear, and (2) the singular vectors of its Jacobian lie in low-dimensional semantic subspaces. We provide a solid theoretical basis to justify the linearity and low-rankness in the PMP. These insights allow us to propose an unsupervised, single-step, training-free LOw-rank COntrollable image editing (LOCO Edit) method for precise local editing in diffusion models. LOCO Edit identified editing directions with nice properties: homogeneity, transferability, composability, and linearity. These properties of LOCO Edit benefit greatly from the low-dimensional semantic subspace. Our method can further be extended to unsupervised or text-supervised editing in various text-to-image diffusion models (T-LOCO Edit). Finally, extensive empirical experiments demonstrate the effectiveness and efficiency of LOCO Edit. The codes will be released at https://github.com/ChicyChen/LOCO-Edit.
We propose an image-conditioned diffusion model to estimate high angular resolution diffusion weighted imaging (DWI) from a low angular resolution acquisition. Our model, which we call QID$^2$, takes as input a set of low angular resolution DWI data and uses this information to estimate the DWI data associated with a target gradient direction. We leverage a U-Net architecture with cross-attention to preserve the positional information of the reference images, further guiding the target image generation. We train and evaluate QID$^2$ on single-shell DWI samples curated from the Human Connectome Project (HCP) dataset. Specifically, we sub-sample the HCP gradient directions to produce low angular resolution DWI data and train QID$^2$ to reconstruct the missing high angular resolution samples. We compare QID$^2$ with two state-of-the-art GAN models. Our results demonstrate that QID$^2$ not only achieves higher-quality generated images, but it consistently outperforms the GAN models in downstream tensor estimation across multiple metrics. Taken together, this study highlights the potential of diffusion models, and QID$^2$ in particular, for q-space up-sampling, thus offering a promising toolkit for clinical and research applications.
Fashion image editing is a crucial tool for designers to convey their creative ideas by visualizing design concepts interactively. Current fashion image editing techniques, though advanced with multimodal prompts and powerful diffusion models, often struggle to accurately identify editing regions and preserve the desired garment texture detail. To address these challenges, we introduce a new multimodal fashion image editing architecture based on latent diffusion models, called Detail-Preserved Diffusion Models (DPDEdit). DPDEdit guides the fashion image generation of diffusion models by integrating text prompts, region masks, human pose images, and garment texture images. To precisely locate the editing region, we first introduce Grounded-SAM to predict the editing region based on the user's textual description, and then combine it with other conditions to perform local editing. To transfer the detail of the given garment texture into the target fashion image, we propose a texture injection and refinement mechanism. Specifically, this mechanism employs a decoupled cross-attention layer to integrate textual descriptions and texture images, and incorporates an auxiliary U-Net to preserve the high-frequency details of generated garment texture. Additionally, we extend the VITON-HD dataset using a multimodal large language model to generate paired samples with texture images and textual descriptions. Extensive experiments show that our DPDEdit outperforms state-of-the-art methods in terms of image fidelity and coherence with the given multimodal inputs.
We explore Bird's-Eye View (BEV) generation, converting a BEV map into its corresponding multi-view street images. Valued for its unified spatial representation aiding multi-sensor fusion, BEV is pivotal for various autonomous driving applications. Creating accurate street-view images from BEV maps is essential for portraying complex traffic scenarios and enhancing driving algorithms. Concurrently, diffusion-based conditional image generation models have demonstrated remarkable outcomes, adept at producing diverse, high-quality, and condition-aligned results. Nonetheless, the training of these models demands substantial data and computational resources. Hence, exploring methods to fine-tune these advanced models, like Stable Diffusion, for specific conditional generation tasks emerges as a promising avenue. In this paper, we introduce a practical framework for generating images from a BEV layout. Our approach comprises two main components: the Neural View Transformation and the Street Image Generation. The Neural View Transformation phase converts the BEV map into aligned multi-view semantic segmentation maps by learning the shape correspondence between the BEV and perspective views. Subsequently, the Street Image Generation phase utilizes these segmentations as a condition to guide a fine-tuned latent diffusion model. This finetuning process ensures both view and style consistency. Our model leverages the generative capacity of large pretrained diffusion models within traffic contexts, effectively yielding diverse and condition-coherent street view images.
We introduce $\texttt{ReMOVE}$, a novel reference-free metric for assessing object erasure efficacy in diffusion-based image editing models post-generation. Unlike existing measures such as LPIPS and CLIPScore, $\texttt{ReMOVE}$ addresses the challenge of evaluating inpainting without a reference image, common in practical scenarios. It effectively distinguishes between object removal and replacement. This is a key issue in diffusion models due to stochastic nature of image generation. Traditional metrics fail to align with the intuitive definition of inpainting, which aims for (1) seamless object removal within masked regions (2) while preserving the background continuity. $\texttt{ReMOVE}$ not only correlates with state-of-the-art metrics and aligns with human perception but also captures the nuanced aspects of the inpainting process, providing a finer-grained evaluation of the generated outputs.
We introduce Kvasir-VQA, an extended dataset derived from the HyperKvasir and Kvasir-Instrument datasets, augmented with question-and-answer annotations to facilitate advanced machine learning tasks in Gastrointestinal (GI) diagnostics. This dataset comprises 6,500 annotated images spanning various GI tract conditions and surgical instruments, and it supports multiple question types including yes/no, choice, location, and numerical count. The dataset is intended for applications such as image captioning, Visual Question Answering (VQA), text-based generation of synthetic medical images, object detection, and classification. Our experiments demonstrate the dataset's effectiveness in training models for three selected tasks, showcasing significant applications in medical image analysis and diagnostics. We also present evaluation metrics for each task, highlighting the usability and versatility of our dataset. The dataset and supporting artifacts are available at https://datasets.simula.no/kvasir-vqa.