Person image synthesis with controllable body poses and appearances is an essential task owing to the practical needs in the context of virtual try-on, image editing and video production. However, existing methods face significant challenges with details missing, limbs distortion and the garment style deviation. To address these issues, we propose a Disentangled Representations Diffusion Model (DRDM) to generate photo-realistic images from source portraits in specific desired poses and appearances. First, a pose encoder is responsible for encoding pose features into a high-dimensional space to guide the generation of person images. Second, a body-part subspace decoupling block (BSDB) disentangles features from the different body parts of a source figure and feeds them to the various layers of the noise prediction block, thereby supplying the network with rich disentangled features for generating a realistic target image. Moreover, during inference, we develop a parsing map-based disentangled classifier-free guided sampling method, which amplifies the conditional signals of texture and pose. Extensive experimental results on the Deepfashion dataset demonstrate the effectiveness of our approach in achieving pose transfer and appearance control.
Diffusion-based text-to-image models have shown immense potential for various image-related tasks. However, despite their prominence and popularity, customizing these models using unauthorized data also brings serious privacy and intellectual property issues. Existing methods introduce protective perturbations based on adversarial attacks, which are applied to the customization samples. In this systematization of knowledge, we present a comprehensive survey of protective perturbation methods designed to prevent unauthorized data usage in diffusion-based image generation. We establish the threat model and categorize the downstream tasks relevant to these methods, providing a detailed analysis of their designs. We also propose a completed evaluation framework for these perturbation techniques, aiming to advance research in this field.
Instance features in images exhibit spurious correlations with background features, affecting the training process of deep neural classifiers. This leads to insufficient attention to instance features by the classifier, resulting in erroneous classification outcomes. In this paper, we propose a data augmentation method called Spurious Correlations Guided Synthesis (SCGS) that mitigates spurious correlations through image generation model. This approach does not require expensive spurious attribute (group) labels for the training data and can be widely applied to other debiasing methods. Specifically, SCGS first identifies the incorrect attention regions of a pre-trained classifier on the training images, and then uses an image generation model to generate new training data based on these incorrect attended regions. SCGS increases the diversity and scale of the dataset to reduce the impact of spurious correlations on classifiers. Changes in the classifier's attention regions and experimental results on three different domain datasets demonstrate that this method is effective in reducing the classifier's reliance on spurious correlations.
We present 1.58-bit FLUX, the first successful approach to quantizing the state-of-the-art text-to-image generation model, FLUX.1-dev, using 1.58-bit weights (i.e., values in {-1, 0, +1}) while maintaining comparable performance for generating 1024 x 1024 images. Notably, our quantization method operates without access to image data, relying solely on self-supervision from the FLUX.1-dev model. Additionally, we develop a custom kernel optimized for 1.58-bit operations, achieving a 7.7x reduction in model storage, a 5.1x reduction in inference memory, and improved inference latency. Extensive evaluations on the GenEval and T2I Compbench benchmarks demonstrate the effectiveness of 1.58-bit FLUX in maintaining generation quality while significantly enhancing computational efficiency.
Recent research in subject-driven generation increasingly emphasizes the importance of selective subject features. Nevertheless, accurately selecting the content in a given reference image still poses challenges, especially when selecting the similar subjects in an image (e.g., two different dogs). Some methods attempt to use text prompts or pixel masks to isolate specific elements. However, text prompts often fall short in precisely describing specific content, and pixel masks are often expensive. To address this, we introduce P3S-Diffusion, a novel architecture designed for context-selected subject-driven generation via point supervision. P3S-Diffusion leverages minimal cost label (e.g., points) to generate subject-driven images. During fine-tuning, it can generate an expanded base mask from these points, obviating the need for additional segmentation models. The mask is employed for inpainting and aligning with subject representation. The P3S-Diffusion preserves fine features of the subjects through Multi-layers Condition Injection. Enhanced by the Attention Consistency Loss for improved training, extensive experiments demonstrate its excellent feature preservation and image generation capabilities.
The in-image machine translation task involves translating text embedded within images, with the translated results presented in image format. While this task has numerous applications in various scenarios such as film poster translation and everyday scene image translation, existing methods frequently neglect the aspect of consistency throughout this process. We propose the need to uphold two types of consistency in this task: translation consistency and image generation consistency. The former entails incorporating image information during translation, while the latter involves maintaining consistency between the style of the text-image and the original image, ensuring background integrity. To address these consistency requirements, we introduce a novel two-stage framework named HCIIT (High-Consistency In-Image Translation) which involves text-image translation using a multimodal multilingual large language model in the first stage and image backfilling with a diffusion model in the second stage. Chain of thought learning is utilized in the first stage to enhance the model's ability to leverage image information during translation. Subsequently, a diffusion model trained for style-consistent text-image generation ensures uniformity in text style within images and preserves background details. A dataset comprising 400,000 style-consistent pseudo text-image pairs is curated for model training. Results obtained on both curated test sets and authentic image test sets validate the effectiveness of our framework in ensuring consistency and producing high-quality translated images.
Image generation in the fashion domain has predominantly focused on preserving body characteristics or following input prompts, but little attention has been paid to improving the inherent fashionability of the output images. This paper presents a novel diffusion model-based approach that generates fashion images with improved fashionability while maintaining control over key attributes. Key components of our method include: 1) fashionability enhancement, which ensures that the generated images are more fashionable than the input; 2) preservation of body characteristics, encouraging the generated images to maintain the original shape and proportions of the input; and 3) automatic fashion optimization, which does not rely on manual input or external prompts. We also employ two methods to collect training data for guidance while generating and evaluating the images. In particular, we rate outfit images using fashionability scores annotated by multiple fashion experts through OpenSkill-based and five critical aspect-based pairwise comparisons. These methods provide complementary perspectives for assessing and improving the fashionability of the generated images. The experimental results show that our approach outperforms the baseline Fashion++ in generating images with superior fashionability, demonstrating its effectiveness in producing more stylish and appealing fashion images.
Recent vision-language foundation models still frequently produce outputs misaligned with their inputs, evidenced by object hallucination in captioning and prompt misalignment in the text-to-image generation model. Recent studies have explored methods for identifying misaligned elements, aiming not only to enhance interpretability but also to improve model performance. However, current approaches primarily rely on large foundation models in a zero-shot manner or fine-tuned models with human annotations, which limits scalability due to significant computational costs. This work proposes a novel approach, dubbed CLIP4DM, for detecting dense misalignments from pre-trained CLIP, specifically focusing on pinpointing misaligned words between image and text. We carefully revamp the gradient-based attribution computation method, enabling negative gradient of individual text tokens to indicate misalignment. We also propose F-CLIPScore, which aggregates misaligned attributions with a global alignment score. We evaluate our method on various dense misalignment detection benchmarks, covering various image and text domains and misalignment types. Our method demonstrates state-of-the-art performance among zero-shot models and competitive performance with fine-tuned models while maintaining superior efficiency. Our qualitative examples show that our method has a unique strength to detect entity-level objects, intangible objects, and attributes that can not be easily detected for existing works. We conduct ablation studies and analyses to highlight the strengths and limitations of our approach. Our code is publicly available at https://github.com/naver-ai/CLIP4DM.
The text-to-image (T2I) personalization diffusion model can generate images of the novel concept based on the user input text caption. However, existing T2I personalized methods either require test-time fine-tuning or fail to generate images that align well with the given text caption. In this work, we propose a new T2I personalization diffusion model, Dense-Face, which can generate face images with a consistent identity as the given reference subject and align well with the text caption. Specifically, we introduce a pose-controllable adapter for the high-fidelity image generation while maintaining the text-based editing ability of the pre-trained stable diffusion (SD). Additionally, we use internal features of the SD UNet to predict dense face annotations, enabling the proposed method to gain domain knowledge in face generation. Empirically, our method achieves state-of-the-art or competitive generation performance in image-text alignment, identity preservation, and pose control.
Recently, Text-to-Image (T2I) generation models have achieved significant advancements. Correspondingly, many automated metrics have emerged to evaluate the image-text alignment capabilities of generative models. However, the performance comparison among these automated metrics is limited by existing small datasets. Additionally, these datasets lack the capacity to assess the performance of automated metrics at a fine-grained level. In this study, we contribute an EvalMuse-40K benchmark, gathering 40K image-text pairs with fine-grained human annotations for image-text alignment-related tasks. In the construction process, we employ various strategies such as balanced prompt sampling and data re-annotation to ensure the diversity and reliability of our benchmark. This allows us to comprehensively evaluate the effectiveness of image-text alignment metrics for T2I models. Meanwhile, we introduce two new methods to evaluate the image-text alignment capabilities of T2I models: FGA-BLIP2 which involves end-to-end fine-tuning of a vision-language model to produce fine-grained image-text alignment scores and PN-VQA which adopts a novel positive-negative VQA manner in VQA models for zero-shot fine-grained evaluation. Both methods achieve impressive performance in image-text alignment evaluations. We also use our methods to rank current AIGC models, in which the results can serve as a reference source for future study and promote the development of T2I generation. The data and code will be made publicly available.
Diffusion Probabilistic Models (DPMs) have emerged as the de facto approach for high-fidelity image synthesis, operating diffusion processes on continuous VAE latent, which significantly differ from the text generation methods employed by Large Language Models (LLMs). In this paper, we introduce a novel generative framework, the Recurrent Diffusion Probabilistic Model (RDPM), which enhances the diffusion process through a recurrent token prediction mechanism, thereby pioneering the field of Discrete Diffusion. By progressively introducing Gaussian noise into the latent representations of images and encoding them into vector-quantized tokens in a recurrent manner, RDPM facilitates a unique diffusion process on discrete-value domains. This process iteratively predicts the token codes for subsequent timesteps, transforming the initial standard Gaussian noise into the source data distribution, aligning with GPT-style models in terms of the loss function. RDPM demonstrates superior performance while benefiting from the speed advantage of requiring only a few inference steps. This model not only leverages the diffusion process to ensure high-quality generation but also converts continuous signals into a series of high-fidelity discrete tokens, thereby maintaining a unified optimization strategy with other discrete tokens, such as text. We anticipate that this work will contribute to the development of a unified model for multimodal generation, specifically by integrating continuous signal domains such as images, videos, and audio with text. We will release the code and model weights to the open-source community.
Active Learning (AL) represents a crucial methodology within machine learning, emphasizing the identification and utilization of the most informative samples for efficient model training. However, a significant challenge of AL is its dependence on the limited labeled data samples and data distribution, resulting in limited performance. To address this limitation, this paper integrates the zero-shot text-to-image (T2I) synthesis and active learning by designing a novel framework that can efficiently train a machine learning (ML) model sorely using the text description. Specifically, we leverage the AL criteria to optimize the text inputs for generating more informative and diverse data samples, annotated by the pseudo-label crafted from text, then served as a synthetic dataset for active learning. This approach reduces the cost of data collection and annotation while increasing the efficiency of model training by providing informative training samples, enabling a novel end-to-end ML task from text description to vision models. Through comprehensive evaluations, our framework demonstrates consistent and significant improvements over traditional AL methods.
Surveys and interviews are widely used for collecting insights on emerging or hypothetical scenarios. Traditional human-led methods often face challenges related to cost, scalability, and consistency. Recently, various domains have begun to explore the use of conversational agents (chatbots) powered by generative artificial intelligence (AI) technologies. However, considering decisions in transportation investments and policies often carry significant public and environmental stakes, surveys and interviews face unique challenges in integrating AI agents, underscoring the need for a rigorous, resource-efficient approach that enhances participant engagement and ensures privacy. This paper addresses this gap by introducing a modular approach and its resulting parameterized process for designing AI agents. We detail the system architecture, integrating engineered prompts, specialized knowledge bases, and customizable, goal-oriented conversational logic. We demonstrate the adaptability, generalizability, and efficacy of our modular approach through three empirical studies: (1) travel preference surveys, highlighting conditional logic and multimodal (voice, text, and image generation) capabilities; (2) public opinion elicitation on a newly constructed, novel infrastructure project, showcasing question customization and multilingual (English and French) capabilities; and (3) expert consultation about the impact of technologies on future transportation systems, highlighting real-time, clarification request capabilities for open-ended questions, resilience in handling erratic inputs, and efficient transcript postprocessing. The results suggest that the AI agent increases completion rates and response quality. Furthermore, the modular approach demonstrates controllability, flexibility, and robustness while addressing key ethical, privacy, security, and token consumption concerns.
To efficiently adapt large models or to train generative models of neural representations, Hypernetworks have drawn interest. While hypernetworks work well, training them is cumbersome, and often requires ground truth optimized weights for each sample. However, obtaining each of these weights is a training problem of its own-one needs to train, e.g., adaptation weights or even an entire neural field for hypernetworks to regress to. In this work, we propose a method to train hypernetworks, without the need for any per-sample ground truth. Our key idea is to learn a Hypernetwork `Field` and estimate the entire trajectory of network weight training instead of simply its converged state. In other words, we introduce an additional input to the Hypernetwork, the convergence state, which then makes it act as a neural field that models the entire convergence pathway of a task network. A critical benefit in doing so is that the gradient of the estimated weights at any convergence state must then match the gradients of the original task -- this constraint alone is sufficient to train the Hypernetwork Field. We demonstrate the effectiveness of our method through the task of personalized image generation and 3D shape reconstruction from images and point clouds, demonstrating competitive results without any per-sample ground truth.
Data scarcity in medical imaging poses significant challenges due to privacy concerns. Diffusion models, a recent generative modeling technique, offer a potential solution by generating synthetic and realistic data. However, questions remain about the performance of convolutional neural network (CNN) models on original and synthetic datasets. If diffusion-generated samples can help CNN models perform comparably to those trained on original datasets, reliance on patient-specific data for training CNNs might be reduced. In this study, we investigated the effectiveness of diffusion models for generating synthetic medical images to train CNNs in three domains: Brain Tumor MRI, Acute Lymphoblastic Leukemia (ALL), and SARS-CoV-2 CT scans. A diffusion model was trained to generate synthetic datasets for each domain. Pre-trained CNN architectures were then trained on these synthetic datasets and evaluated on unseen real data. All three datasets achieved promising classification performance using CNNs trained on synthetic data. Local Interpretable Model-Agnostic Explanations (LIME) analysis revealed that the models focused on relevant image features for classification. This study demonstrates the potential of diffusion models to generate synthetic medical images for training CNNs in medical image analysis.
Recently, the success of text-to-image synthesis has greatly advanced the development of identity customization techniques, whose main goal is to produce realistic identity-specific photographs based on text prompts and reference face images. However, it is difficult for existing identity customization methods to simultaneously meet the various requirements of different real-world applications, including the identity fidelity of small face, the control of face location, pose and expression, as well as the customization of multiple persons. To this end, we propose a scale-robust and fine-controllable method, namely RealisID, which learns different control capabilities through the cooperation between a pair of local and global branches. Specifically, by using cropping and up-sampling operations to filter out face-irrelevant information, the local branch concentrates the fine control of facial details and the scale-robust identity fidelity within the face region. Meanwhile, the global branch manages the overall harmony of the entire image. It also controls the face location by taking the location guidance as input. As a result, RealisID can benefit from the complementarity of these two branches. Finally, by implementing our branches with two different variants of ControlNet, our method can be easily extended to handle multi-person customization, even only trained on single-person datasets. Extensive experiments and ablation studies indicate the effectiveness of RealisID and verify its ability in fulfilling all the requirements mentioned above.
Current image generation methods, such as latent diffusion and discrete token-based generation, depend on a two-stage training approach. In stage 1, an auto-encoder is trained to compress an image into a latent space; in stage 2, a generative model is trained to learn a distribution over that latent space. Most work focuses on maximizing stage 1 performance independent of stage 2, assuming better reconstruction always leads to better generation. However, we show this is not strictly true. Smaller stage 2 models can benefit from more compressed stage 1 latents even if reconstruction performance worsens, showing a fundamental trade-off between compression and generation modeling capacity. To better optimize this trade-off, we introduce Causally Regularized Tokenization (CRT), which uses knowledge of the stage 2 generation modeling procedure to embed useful inductive biases in stage 1 latents. This regularization makes stage 1 reconstruction performance worse, but makes stage 2 generation performance better by making the tokens easier to model: we are able to improve compute efficiency 2-3$\times$ over baseline and match state-of-the-art discrete autoregressive ImageNet generation (2.18 FID) with less than half the tokens per image (256 vs. 576) and a fourth the total model parameters (775M vs. 3.1B) as the previous SOTA (LlamaGen).
The personalization model has gained significant attention in image generation yet remains underexplored for large vision-language models (LVLMs). Beyond generic ones, with personalization, LVLMs handle interactive dialogues using referential concepts (e.g., ``Mike and Susan are talking.'') instead of the generic form (e.g., ``a boy and a girl are talking.''), making the conversation more customizable and referentially friendly. In addition, PLVM is equipped to continuously add new concepts during a dialogue without incurring additional costs, which significantly enhances the practicality. PLVM proposes Aligner, a pre-trained visual encoder to align referential concepts with the queried images. During the dialogues, it extracts features of reference images with these corresponding concepts and recognizes them in the queried image, enabling personalization. We note that the computational cost and parameter count of the Aligner are negligible within the entire framework. With comprehensive qualitative and quantitative analyses, we reveal the effectiveness and superiority of PLVM.
Generative Zero-Shot Learning (ZSL) methods synthesize class-related features based on predefined class semantic prototypes, showcasing superior performance. However, this feature generation paradigm falls short of providing interpretable insights. In addition, existing approaches rely on semantic prototypes annotated by human experts, which exhibit a significant limitation in their scalability to generalized scenes. To overcome these deficiencies, a natural solution is to generate images for unseen classes using text prompts. To this end, We present DIG-ZSL, a novel Discriminative Image Generation framework for Zero-Shot Learning. Specifically, to ensure the generation of discriminative images for training an effective ZSL classifier, we learn a discriminative class token (DCT) for each unseen class under the guidance of a pre-trained category discrimination model (CDM). Harnessing DCTs, we can generate diverse and high-quality images, which serve as informative unseen samples for ZSL tasks. In this paper, the extensive experiments and visualizations on four datasets show that our DIG-ZSL: (1) generates diverse and high-quality images, (2) outperforms previous state-of-the-art nonhuman-annotated semantic prototype-based methods by a large margin, and (3) achieves comparable or better performance than baselines that leverage human-annotated semantic prototypes. The codes will be made available upon acceptance of the paper.
Autoregressive (AR) models have achieved state-of-the-art performance in text and image generation but suffer from slow generation due to the token-by-token process. We ask an ambitious question: can a pre-trained AR model be adapted to generate outputs in just one or two steps? If successful, this would significantly advance the development and deployment of AR models. We notice that existing works that try to speed up AR generation by generating multiple tokens at once fundamentally cannot capture the output distribution due to the conditional dependencies between tokens, limiting their effectiveness for few-step generation. To address this, we propose Distilled Decoding (DD), which uses flow matching to create a deterministic mapping from Gaussian distribution to the output distribution of the pre-trained AR model. We then train a network to distill this mapping, enabling few-step generation. DD doesn't need the training data of the original AR model, making it more practical. We evaluate DD on state-of-the-art image AR models and present promising results on ImageNet-256. For VAR, which requires 10-step generation, DD enables one-step generation (6.3$\times$ speed-up), with an acceptable increase in FID from 4.19 to 9.96. For LlamaGen, DD reduces generation from 256 steps to 1, achieving an 217.8$\times$ speed-up with a comparable FID increase from 4.11 to 11.35. In both cases, baseline methods completely fail with FID>100. DD also excels on text-to-image generation, reducing the generation from 256 steps to 2 for LlamaGen with minimal FID increase from 25.70 to 28.95. As the first work to demonstrate the possibility of one-step generation for image AR models, DD challenges the prevailing notion that AR models are inherently slow, and opens up new opportunities for efficient AR generation. The project website is at https://imagination-research.github.io/distilled-decoding.