There exists recent work in computer vision, named VAR, that proposes a new autoregressive paradigm for image generation. Diverging from the vanilla next-token prediction, VAR structurally reformulates the image generation into a coarse to fine next-scale prediction. In this paper, we show that this scale-wise autoregressive framework can be effectively decoupled into \textit{intra-scale modeling}, which captures local spatial dependencies within each scale, and \textit{inter-scale modeling}, which models cross-scale relationships progressively from coarse-to-fine scales. This decoupling structure allows to rebuild VAR in a more computationally efficient manner. Specifically, for intra-scale modeling -- crucial for generating high-fidelity images -- we retain the original bidirectional self-attention design to ensure comprehensive modeling; for inter-scale modeling, which semantically connects different scales but is computationally intensive, we apply linear-complexity mechanisms like Mamba to substantially reduce computational overhead. We term this new framework M-VAR. Extensive experiments demonstrate that our method outperforms existing models in both image quality and generation speed. For example, our 1.5B model, with fewer parameters and faster inference speed, outperforms the largest VAR-d30-2B. Moreover, our largest model M-VAR-d32 impressively registers 1.78 FID on ImageNet 256$\times$256 and outperforms the prior-art autoregressive models LlamaGen/VAR by 0.4/0.19 and popular diffusion models LDM/DiT by 1.82/0.49, respectively. Code is avaiable at \url{https://github.com/OliverRensu/MVAR}.
Text-to-image generation and text-guided image manipulation have received considerable attention in the field of image generation tasks. However, the mainstream evaluation methods for these tasks have difficulty in evaluating whether all the information from the input text is accurately reflected in the generated images, and they mainly focus on evaluating the overall alignment between the input text and the generated images. This paper proposes new evaluation metrics that assess the alignment between input text and generated images for every individual object. Firstly, according to the input text, chatGPT is utilized to produce questions for the generated images. After that, we use Visual Question Answering(VQA) to measure the relevance of the generated images to the input text, which allows for a more detailed evaluation of the alignment compared to existing methods. In addition, we use Non-Reference Image Quality Assessment(NR-IQA) to evaluate not only the text-image alignment but also the quality of the generated images. Experimental results show that our proposed evaluation approach is the superior metric that can simultaneously assess finer text-image alignment and image quality while allowing for the adjustment of these ratios.
In recent years, image synthesis has achieved remarkable advancements, enabling diverse applications in content creation, virtual reality, and beyond. We introduce a novel approach to image generation using Auto-Regressive (AR) modeling, which leverages a next-detail prediction strategy for enhanced fidelity and scalability. While AR models have achieved transformative success in language modeling, replicating this success in vision tasks has presented unique challenges due to the inherent spatial dependencies in images. Our proposed method addresses these challenges by iteratively adding finer details to an image compositionally, constructing it as a hierarchical combination of base and detail image factors. This strategy is shown to be more effective than the conventional next-token prediction and even surpasses the state-of-the-art next-scale prediction approaches. A key advantage of this method is its scalability to higher resolutions without requiring full model retraining, making it a versatile solution for high-resolution image generation.
As a highly expressive generative model, diffusion models have demonstrated exceptional success across various domains, including image generation, natural language processing, and combinatorial optimization. However, as data distributions grow more complex, training these models to convergence becomes increasingly computationally intensive. While diffusion models are typically trained using uniform timestep sampling, our research shows that the variance in stochastic gradients varies significantly across timesteps, with high-variance timesteps becoming bottlenecks that hinder faster convergence. To address this issue, we introduce a non-uniform timestep sampling method that prioritizes these more critical timesteps. Our method tracks the impact of gradient updates on the objective for each timestep, adaptively selecting those most likely to minimize the objective effectively. Experimental results demonstrate that this approach not only accelerates the training process, but also leads to improved performance at convergence. Furthermore, our method shows robust performance across various datasets, scheduling strategies, and diffusion architectures, outperforming previously proposed timestep sampling and weighting heuristics that lack this degree of robustness.
In recent years, text-to-image (T2I) generation models have made significant progress in generating high-quality images that align with text descriptions. However, these models also face the risk of unsafe generation, potentially producing harmful content that violates usage policies, such as explicit material. Existing safe generation methods typically focus on suppressing inappropriate content by erasing undesired concepts from visual representations, while neglecting to sanitize the textual representation. Although these methods help mitigate the risk of misuse to some extent, their robustness remains insufficient when dealing with adversarial attacks. Given that semantic consistency between input text and output image is a core requirement of T2I models, we identify that textual representations are likely the primary source of unsafe generation. To this end, we propose Embedding Sanitizer (ES), which enhances the safety of T2I models by sanitizing inappropriate concepts in prompt embeddings. To our knowledge, ES is the first interpretable safe generation framework that assigns a score to each token in the prompt to indicate its potential harmfulness. In addition, ES adopts a plug-and-play modular design, offering compatibility for seamless integration with various T2I models and other safeguards. Evaluations on five prompt benchmarks show that ES outperforms eleven existing safeguard baselines, achieving state-of-the-art robustness while maintaining high-quality image generation.
Remarkable progress has been achieved in image generation with the introduction of generative models. However, precisely controlling the content in generated images remains a challenging task due to their fundamental training objective. This paper addresses this challenge by proposing a novel image generation framework explicitly designed to incorporate desired content in output images. The framework utilizes advanced encoding techniques, integrating subnetworks called content fusion and frequency encoding modules. The frequency encoding module first captures features and structures of reference images by exclusively focusing on selected frequency components. Subsequently, the content fusion module generates a content-guiding vector that encapsulates desired content features. During the image generation process, content-guiding vectors from real images are fused with projected noise vectors. This ensures the production of generated images that not only maintain consistent content from guiding images but also exhibit diverse stylistic variations. To validate the effectiveness of the proposed framework in preserving content attributes, extensive experiments are conducted on widely used benchmark datasets, including Flickr-Faces-High Quality, Animal Faces High Quality, and Large-scale Scene Understanding datasets.
We introduce Mediffusion -- a new method for semi-supervised learning with explainable classification based on a joint diffusion model. The medical imaging domain faces unique challenges due to scarce data labelling -- insufficient for standard training, and critical nature of the applications that require high performance, confidence, and explainability of the models. In this work, we propose to tackle those challenges with a single model that combines standard classification with a diffusion-based generative task in a single shared parametrisation. By sharing representations, our model effectively learns from both labeled and unlabeled data while at the same time providing accurate explanations through counterfactual examples. In our experiments, we show that our Mediffusion achieves results comparable to recent semi-supervised methods while providing more reliable and precise explanations.
Diffusion models have revitalized the image generation domain, playing crucial roles in both academic research and artistic expression. With the emergence of new diffusion models, assessing the performance of text-to-image models has become increasingly important. Current metrics focus on directly matching the input text with the generated image, but due to cross-modal information asymmetry, this leads to unreliable or incomplete assessment results. Motivated by this, we introduce the Image Regeneration task in this study to assess text-to-image models by tasking the T2I model with generating an image according to the reference image. We use GPT4V to bridge the gap between the reference image and the text input for the T2I model, allowing T2I models to understand image content. This evaluation process is simplified as comparisons between the generated image and the reference image are straightforward. Two regeneration datasets spanning content-diverse and style-diverse evaluation dataset are introduced to evaluate the leading diffusion models currently available. Additionally, we present ImageRepainter framework to enhance the quality of generated images by improving content comprehension via MLLM guided iterative generation and revision. Our comprehensive experiments have showcased the effectiveness of this framework in assessing the generative capabilities of models. By leveraging MLLM, we have demonstrated that a robust T2M can produce images more closely resembling the reference image.
Recent advances in image generation, particularly via diffusion models, have led to impressive improvements in image synthesis quality. Despite this, diffusion models are still challenged by model-induced artifacts and limited stability in image fidelity. In this work, we hypothesize that the primary cause of this issue is the improper resampling operation that introduces aliasing in the diffusion model and a careful alias-free resampling dictated by image processing theory can improve the model's performance in image synthesis. We propose the integration of alias-free resampling layers into the UNet architecture of diffusion models without adding extra trainable parameters, thereby maintaining computational efficiency. We then assess whether these theory-driven modifications enhance image quality and rotational equivariance. Our experimental results on benchmark datasets, including CIFAR-10, MNIST, and MNIST-M, reveal consistent gains in image quality, particularly in terms of FID and KID scores. Furthermore, we propose a modified diffusion process that enables user-controlled rotation of generated images without requiring additional training. Our findings highlight the potential of theory-driven enhancements such as alias-free resampling in generative models to improve image quality while maintaining model efficiency and pioneer future research directions to incorporate them into video-generating diffusion models, enabling deeper exploration of the applications of alias-free resampling in generative modeling.
Diffusion models have recently emerged as a potent tool in generative modeling. However, their inherent iterative nature often results in sluggish image generation due to the requirement for multiple model evaluations. Recent progress has unveiled the intrinsic link between diffusion models and Probability Flow Ordinary Differential Equations (ODEs), thus enabling us to conceptualize diffusion models as ODE systems. Simultaneously, Physics Informed Neural Networks (PINNs) have substantiated their effectiveness in solving intricate differential equations through implicit modeling of their solutions. Building upon these foundational insights, we introduce Physics Informed Distillation (PID), which employs a student model to represent the solution of the ODE system corresponding to the teacher diffusion model, akin to the principles employed in PINNs. Through experiments on CIFAR 10 and ImageNet 64x64, we observe that PID achieves performance comparable to recent distillation methods. Notably, it demonstrates predictable trends concerning method-specific hyperparameters and eliminates the need for synthetic dataset generation during the distillation process. Both of which contribute to its easy-to-use nature as a distillation approach for Diffusion Models. Our code and pre-trained checkpoint are publicly available at: https://github.com/pantheon5100/pid_diffusion.git.
Diffusion Transformers (DiTs) have recently achieved remarkable success in text-guided image generation. In image editing, DiTs project text and image inputs to a joint latent space, from which they decode and synthesize new images. However, it remains largely unexplored how multimodal information collectively forms this joint space and how they guide the semantics of the synthesized images. In this paper, we investigate the latent space of DiT models and uncover two key properties: First, DiT's latent space is inherently semantically disentangled, where different semantic attributes can be controlled by specific editing directions. Second, consistent semantic editing requires utilizing the entire joint latent space, as neither encoded image nor text alone contains enough semantic information. We show that these editing directions can be obtained directly from text prompts, enabling precise semantic control without additional training or mask annotations. Based on these insights, we propose a simple yet effective Encode-Identify-Manipulate (EIM) framework for zero-shot fine-grained image editing. Specifically, we first encode both the given source image and the text prompt that describes the image, to obtain the joint latent embedding. Then, using our proposed Hessian Score Distillation Sampling (HSDS) method, we identify editing directions that control specific target attributes while preserving other image features. These directions are guided by text prompts and used to manipulate the latent embeddings. Moreover, we propose a new metric to quantify the disentanglement degree of the latent space of diffusion models. Extensive experiment results on our new curated benchmark dataset and analysis demonstrate DiT's disentanglement properties and effectiveness of the EIM framework.
TIPO (Text to Image with text pre-sampling for Prompt Optimization) is an innovative framework designed to enhance text-to-image (T2I) generation by language model (LM) for automatic prompt engineering. By refining and extending user-provided prompts, TIPO bridges the gap between simple inputs and the detailed prompts required for high-quality image generation. Unlike previous approaches that rely on Large Language Models (LLMs) or reinforcement learning (RL), TIPO adjusts user input prompts with the distribution of a trained prompt dataset, eliminating the need for complex runtime cost via lightweight model. This pre-sampling approach enables efficient and scalable prompt optimization, grounded in the model's training distribution. Experimental results demonstrate TIPO's effectiveness in improving aesthetic scores, reducing image corruption, and better aligning generated images with dataset distributions. These findings highlight the critical role of prompt engineering in T2I systems and open avenues for broader applications of automatic prompt refinement.
Autoregressive models have demonstrated great performance in natural language processing (NLP) with impressive scalability, adaptability and generalizability. Inspired by their notable success in NLP field, autoregressive models have been intensively investigated recently for computer vision, which perform next-token predictions by representing visual data as visual tokens and enables autoregressive modelling for a wide range of vision tasks, ranging from visual generation and visual understanding to the very recent multimodal generation that unifies visual generation and understanding with a single autoregressive model. This paper provides a systematic review of vision autoregressive models, including the development of a taxonomy of existing methods and highlighting their major contributions, strengths, and limitations, covering various vision tasks such as image generation, video generation, image editing, motion generation, medical image analysis, 3D generation, robotic manipulation, unified multimodal generation, etc. Besides, we investigate and analyze the latest advancements in autoregressive models, including thorough benchmarking and discussion of existing methods across various evaluation datasets. Finally, we outline key challenges and promising directions for future research, offering a roadmap to guide further advancements in vision autoregressive models.
The remarkable progress in neural-network-driven visual data generation, especially with neural rendering techniques like Neural Radiance Fields and 3D Gaussian splatting, offers a powerful alternative to GANs and diffusion models. These methods can produce high-fidelity images and lifelike avatars, highlighting the need for robust detection methods. In response, an unsupervised training technique is proposed that enables the model to extract comprehensive features from the Fourier spectrum magnitude, thereby overcoming the challenges of reconstructing the spectrum due to its centrosymmetric properties. By leveraging the spectral domain and dynamically combining it with spatial domain information, we create a robust multimodal detector that demonstrates superior generalization capabilities in identifying challenging synthetic images generated by the latest image synthesis techniques. To address the absence of a 3D neural rendering-based fake image database, we develop a comprehensive database that includes images generated by diverse neural rendering techniques, providing a robust foundation for evaluating and advancing detection methods.
This paper proposes a process for a classification model for the facial expressions. The proposed process would aid in specific categorisation of children's emotions from 2 emotions namely 'Happy' and 'Sad'. Since the existing emotion recognition systems algorithms primarily train on adult faces, the model developed is achieved by using advanced concepts of models with Squeeze-andExcitation blocks, Convolutional Block Attention modules, and robust data augmentation. Stable Diffusion image synthesis was used for expanding and diversifying the data set generating realistic and various training samples. The model designed using Batch Normalisation, Dropout, and SE Attention mechanisms for the classification of children's emotions achieved an accuracy rate of 89\% due to these methods improving the precision of emotion recognition in children. The relative importance of this issue is raised in this study with an emphasis on the call for a more specific model in emotion detection systems for the young generation with specific direction on how the young people can be assisted to manage emotions while online.
Understanding spatial relations is a crucial cognitive ability for both humans and AI. While current research has predominantly focused on the benchmarking of text-to-image (T2I) models, we propose a more comprehensive evaluation that includes \textit{both} T2I and Large Language Models (LLMs). As spatial relations are naturally understood in a visuo-spatial manner, we develop an approach to convert LLM outputs into an image, thereby allowing us to evaluate both T2I models and LLMs \textit{visually}. We examined the spatial relation understanding of 8 prominent generative models (3 T2I models and 5 LLMs) on a set of 10 common prepositions, as well as assess the feasibility of automatic evaluation methods. Surprisingly, we found that T2I models only achieve subpar performance despite their impressive general image-generation abilities. Even more surprisingly, our results show that LLMs are significantly more accurate than T2I models in generating spatial relations, despite being primarily trained on textual data. We examined reasons for model failures and highlight gaps that can be filled to enable more spatially faithful generations.
Flow diffusion models (FDMs) have recently shown potential in generation tasks due to the high generation quality. However, the current ordinary differential equation (ODE) solver for FDMs, e.g., the Euler solver, still suffers from slow generation since ODE solvers need many number function evaluations (NFE) to keep high-quality generation. In this paper, we propose a novel training-free flow-solver to reduce NFE while maintaining high-quality generation. The key insight for the flow-solver is to leverage the previous steps to reduce the NFE, where a cache is created to reuse these results from the previous steps. Specifically, the Taylor expansion is first used to approximate the ODE. To calculate the high-order derivatives of Taylor expansion, the flow-solver proposes to use the previous steps and a polynomial interpolation to approximate it, where the number of orders we could approximate equals the number of previous steps we cached. We also prove that the flow-solver has a more minor approximation error and faster generation speed. Experimental results on the CIFAR-10, CelebA-HQ, LSUN-Bedroom, LSUN-Church, ImageNet, and real text-to-image generation prove the efficiency of the flow-solver. Specifically, the flow-solver improves the FID-30K from 13.79 to 6.75, from 46.64 to 19.49 with $\text{NFE}=10$ on CIFAR-10 and LSUN-Church, respectively.
Diffusion models have revolutionized image synthesis, garnering significant research interest in recent years. Diffusion is an iterative algorithm in which samples are generated step-by-step, starting from pure noise. This process introduces the notion of diffusion trajectories, i.e., paths from the standard Gaussian distribution to the target image distribution. In this context, we study discriminative algorithms operating on these trajectories. Specifically, given a pre-trained diffusion model, we consider the problem of classifying images as part of the training dataset, generated by the model or originating from an external source. Our approach demonstrates the presence of patterns across steps that can be leveraged for classification. We also conduct ablation studies, which reveal that using higher-order gradient features to characterize the trajectories leads to significant performance gains and more robust algorithms.
Variational autoencoder (VAE) is one of the most common techniques in the field of medical image generation, where this architecture has shown advanced researchers in recent years and has developed into various architectures. VAE has advantages including improving datasets by adding samples in smaller datasets and in datasets with imbalanced classes, and this is how data augmentation works. This paper provides a comprehensive review of studies on VAE in medical imaging, with a special focus on their ability to create synthetic images close to real data so that they can be used for data augmentation. This study reviews important architectures and methods used to develop VAEs for medical images and provides a comparison with other generative models such as GANs on issues such as image quality, and low diversity of generated samples. We discuss recent developments and applications in several medical fields highlighting the ability of VAEs to improve segmentation and classification accuracy.
Vision Transformers and U-Net architectures have been widely adopted in the implementation of Diffusion Models. However, each architecture presents specific challenges while realizing them on-device. Vision Transformers require positional embedding to maintain correspondence between the tokens processed by the transformer, although they offer the advantage of using fixed-size, reusable repetitive blocks following tokenization. The U-Net architecture lacks these attributes, as it utilizes variable-sized intermediate blocks for down-convolution and up-convolution in the noise estimation backbone for the diffusion process. To address these issues, we propose an architecture that utilizes a fixed-size, reusable transformer block as a core structure, making it more suitable for hardware implementation. Our architecture is characterized by low complexity, token-free design, absence of positional embeddings, uniformity, and scalability, making it highly suitable for deployment on mobile and resource-constrained devices. The proposed model exhibit competitive and consistent performance across both unconditional and conditional image generation tasks. The model achieved a state-of-the-art FID score of 1.6 on unconditional image generation with the CelebA.