Large-scale vision-language models, such as CLIP, are known to contain societal bias regarding protected attributes (e.g., gender, age). This paper aims to address the problems of societal bias in CLIP. Although previous studies have proposed to debias societal bias through adversarial learning or test-time projecting, our comprehensive study of these works identifies two critical limitations: 1) loss of attribute information when it is explicitly disclosed in the input and 2) use of the attribute annotations during debiasing process. To mitigate societal bias in CLIP and overcome these limitations simultaneously, we introduce a simple-yet-effective debiasing method called SANER (societal attribute neutralizer) that eliminates attribute information from CLIP text features only of attribute-neutral descriptions. Experimental results show that SANER, which does not require attribute annotations and preserves original information for attribute-specific descriptions, demonstrates superior debiasing ability than the existing methods.
Medical image synthesis generates additional imaging modalities that are costly, invasive or harmful to acquire, which helps to facilitate the clinical workflow. When training pairs are substantially misaligned (e.g., lung MRI-CT pairs with respiratory motion), accurate image synthesis remains a critical challenge. Recent works explored the directional registration module to adjust misalignment in generative adversarial networks (GANs); however, substantial misalignment will lead to 1) suboptimal data mapping caused by correspondence ambiguity, and 2) degraded image fidelity caused by morphology influence on discriminators. To address the challenges, we propose a novel Deformation-aware GAN (DA-GAN) to dynamically correct the misalignment during the image synthesis based on multi-objective inverse consistency. Specifically, in the generative process, three levels of inverse consistency cohesively optimise symmetric registration and image generation for improved correspondence. In the adversarial process, to further improve image fidelity under misalignment, we design deformation-aware discriminators to disentangle the mismatched spatial morphology from the judgement of image fidelity. Experimental results show that DA-GAN achieved superior performance on a public dataset with simulated misalignments and a real-world lung MRI-CT dataset with respiratory motion misalignment. The results indicate the potential for a wide range of medical image synthesis tasks such as radiotherapy planning.
Photo-Sharing Multi-modal dialogue generation requires a dialogue agent not only to generate text responses but also to share photos at the proper moment. Using image text caption as the bridge, a pipeline model integrates an image caption model, a text generation model, and an image generation model to handle this complex multi-modal task. However, representing the images with text captions may loss important visual details and information and cause error propagation in the complex dialogue system. Besides, the pipeline model isolates the three models separately because discrete image text captions hinder end-to-end gradient propagation. We propose the first end-to-end model for photo-sharing multi-modal dialogue generation, which integrates an image perceptron and an image generator with a large language model. The large language model employs the Q-Former to perceive visual images in the input end. For image generation in the output end, we propose a dynamic vocabulary transformation matrix and use straight-through and gumbel-softmax techniques to align the large language model and stable diffusion model and achieve end-to-end gradient propagation. We perform experiments on PhotoChat and DialogCC datasets to evaluate our end-to-end model. Compared with pipeline models, the end-to-end model gains state-of-the-art performances on various metrics of text and image generation. More analysis experiments also verify the effectiveness of the end-to-end model for photo-sharing multi-modal dialogue generation.
Diffusion Probabilistic Models (DPMs) have shown remarkable potential in image generation, but their sampling efficiency is hindered by the need for numerous denoising steps. Most existing solutions accelerate the sampling process by proposing fast ODE solvers. However, the inevitable discretization errors of the ODE solvers are significantly magnified when the number of function evaluations (NFE) is fewer. In this work, we propose PFDiff, a novel training-free and orthogonal timestep-skipping strategy, which enables existing fast ODE solvers to operate with fewer NFE. Specifically, PFDiff initially utilizes gradient replacement from past time steps to predict a "springboard". Subsequently, it employs this "springboard" along with foresight updates inspired by Nesterov momentum to rapidly update current intermediate states. This approach effectively reduces unnecessary NFE while correcting for discretization errors inherent in first-order ODE solvers. Experimental results demonstrate that PFDiff exhibits flexible applicability across various pre-trained DPMs, particularly excelling in conditional DPMs and surpassing previous state-of-the-art training-free methods. For instance, using DDIM as a baseline, we achieved 16.46 FID (4 NFE) compared to 138.81 FID with DDIM on ImageNet 64x64 with classifier guidance, and 13.06 FID (10 NFE) on Stable Diffusion with 7.5 guidance scale.
This paper examines three major generative modelling frameworks: Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and Stable Diffusion models. VAEs are effective at learning latent representations but frequently yield blurry results. GANs can generate realistic images but face issues such as mode collapse. Stable Diffusion models, while producing high-quality images with strong semantic coherence, are demanding in terms of computational resources. Additionally, the paper explores how incorporating Grounding DINO and Grounded SAM with Stable Diffusion improves image accuracy by utilising sophisticated segmentation and inpainting techniques. The analysis guides on selecting suitable models for various applications and highlights areas for further research.
Diffusion models (DMs) have demonstrated great potential in the field of adversarial robustness, where DM-based defense methods can achieve superior defense capability without adversarial training. However, they all require huge computational costs due to the usage of large-scale pre-trained DMs, making it difficult to conduct full evaluation under strong attacks and compare with traditional CNN-based methods. Simply reducing the network size and timesteps in DMs could significantly harm the image generation quality, which invalidates previous frameworks. To alleviate this issue, we redesign the diffusion framework from generating high-quality images to predicting distinguishable image labels. Specifically, we employ an image translation framework to learn many-to-one mapping from input samples to designed orthogonal image labels. Based on this framework, we introduce an efficient Image-to-Image diffusion classifier with a pruned U-Net structure and reduced diffusion timesteps. Besides the framework, we redesign the optimization objective of DMs to fit the target of image classification, where a new classification loss is incorporated in the DM-based image translation framework to distinguish the generated label from those of other classes. We conduct sufficient evaluations of the proposed classifier under various attacks on popular benchmarks. Extensive experiments show that our method achieves better adversarial robustness with fewer computational costs than DM-based and CNN-based methods. The code is available at https://github.com/hfmei/IDC.
Recent work in image and video generation has been adopting the autoregressive LLM architecture due to its generality and potentially easy integration into multi-modal systems. The crux of applying autoregressive training in language generation to visual generation is discretization -- representing continuous data like images and videos as discrete tokens. Common methods of discretizing images and videos include modeling raw pixel values, which are prohibitively lengthy, or vector quantization, which requires convoluted pre-hoc training. In this work, we propose to directly model images and videos as compressed files saved on computers via canonical codecs (e.g., JPEG, AVC/H.264). Using the default Llama architecture without any vision-specific modifications, we pretrain JPEG-LM from scratch to generate images (and AVC-LM to generate videos as a proof of concept), by directly outputting compressed file bytes in JPEG and AVC formats. Evaluation of image generation shows that this simple and straightforward approach is more effective than pixel-based modeling and sophisticated vector quantization baselines (on which our method yields a 31% reduction in FID). Our analysis shows that JPEG-LM has an especial advantage over vector quantization models in generating long-tail visual elements. Overall, we show that using canonical codec representations can help lower the barriers between language generation and visual generation, facilitating future research on multi-modal language/image/video LLMs.
Improvements in diffusion models have boosted the quality of image generation, which has led researchers, companies, and creators to focus on improving watermarking algorithms. This provision would make it possible to clearly identify the creators of generative art. The main challenges that modern watermarking algorithms face have to do with their ability to withstand attacks and encrypt many unique messages, such as user IDs. In this paper, we present METR: Message Enhanced Tree-Ring, which is an approach that aims to address these challenges. METR is built on the Tree-Ring watermarking algorithm, a technique that makes it possible to encode multiple distinct messages without compromising attack resilience or image quality. This ensures the suitability of this watermarking algorithm for any Diffusion Model. In order to surpass the limitations on the quantity of encoded messages, we propose METR++, an enhanced version of METR. This approach, while limited to the Latent Diffusion Model architecture, is designed to inject a virtually unlimited number of unique messages. We demonstrate its robustness to attacks and ability to encrypt many unique messages while preserving image quality, which makes METR and METR++ hold great potential for practical applications in real-world settings. Our code is available at https://github.com/deepvk/metr
Sketch, a powerful artistic technique to capture essential visual information about real-world objects, is increasingly gaining attention in the image synthesis field. However, evaluating the quality of synthesized sketches presents unique unsolved challenges. Current evaluation methods for sketch synthesis are inadequate due to the lack of a unified benchmark dataset, over-reliance on classification accuracy for recognizability, and unfair evaluation of sketches with different levels of simplification. To address these issues, we introduce SketchRef, a benchmark dataset comprising 4 categories of reference photos--animals, human faces, human bodies, and common objects--alongside novel evaluation metrics. Considering that classification accuracy is insufficient to measure the structural consistency between a sketch and its reference photo, we propose the mean Object Keypoint Similarity (mOKS) metric, utilizing pose estimation to assess structure-level recognizability. To ensure fair evaluation sketches with different simplification levels, we propose a recognizability calculation method constrained by simplicity. We also collect 8K responses from art enthusiasts, validating the effectiveness of our proposed evaluation methods. We hope this work can provide a comprehensive evaluation of sketch synthesis algorithms, thereby aligning their performance more closely with human understanding.
We propose in this paper an analytically new construct of a diffusion model whose drift and diffusion parameters yield an exponentially time-decaying Signal to Noise Ratio in the forward process. In reverse, the construct cleverly carries out the learning of the diffusion coefficients on the structure of clean images using an autoencoder. The proposed methodology significantly accelerates the diffusion process, reducing the required diffusion time steps from around 1000 seen in conventional models to 200-500 without compromising image quality in the reverse-time diffusion. In a departure from conventional models which typically use time-consuming multiple runs, we introduce a parallel data-driven model to generate a reverse-time diffusion trajectory in a single run of the model. The resulting collective block-sequential generative model eliminates the need for MCMC-based sub-sampling correction for safeguarding and improving image quality, to further improve the acceleration of image generation. Collectively, these advancements yield a generative model that is an order of magnitude faster than conventional approaches, while maintaining high fidelity and diversity in generated images, hence promising widespread applicability in rapid image synthesis tasks.
Diffusion models excel at capturing the natural design spaces of images, molecules, DNA, RNA, and protein sequences. However, rather than merely generating designs that are natural, we often aim to optimize downstream reward functions while preserving the naturalness of these design spaces. Existing methods for achieving this goal often require ``differentiable'' proxy models (\textit{e.g.}, classifier guidance or DPS) or involve computationally expensive fine-tuning of diffusion models (\textit{e.g.}, classifier-free guidance, RL-based fine-tuning). In our work, we propose a new method to address these challenges. Our algorithm is an iterative sampling method that integrates soft value functions, which looks ahead to how intermediate noisy states lead to high rewards in the future, into the standard inference procedure of pre-trained diffusion models. Notably, our approach avoids fine-tuning generative models and eliminates the need to construct differentiable models. This enables us to (1) directly utilize non-differentiable features/reward feedback, commonly used in many scientific domains, and (2) apply our method to recent discrete diffusion models in a principled way. Finally, we demonstrate the effectiveness of our algorithm across several domains, including image generation, molecule generation, and DNA/RNA sequence generation. The code is available at \href{https://github.com/masa-ue/SVDD}{https://github.com/masa-ue/SVDD}.
Large Language Models (LLMs) have showcased exceptional ability in causal reasoning from textual information. However, will these causalities remain straightforward for Vision Large Language Models (VLLMs) when only visual hints are provided? Motivated by this, we propose a novel Multimodal Causal Reasoning benchmark, namely MuCR, to challenge VLLMs to infer semantic cause-and-effect relationship when solely relying on visual cues such as action, appearance, clothing, and environment. Specifically, we introduce a prompt-driven image synthesis approach to create siamese images with embedded semantic causality and visual cues, which can effectively evaluate VLLMs' causal reasoning capabilities. Additionally, we develop tailored metrics from multiple perspectives, including image-level match, phrase-level understanding, and sentence-level explanation, to comprehensively assess VLLMs' comprehension abilities. Our extensive experiments reveal that the current state-of-the-art VLLMs are not as skilled at multimodal causal reasoning as we might have hoped. Furthermore, we perform a comprehensive analysis to understand these models' shortcomings from different views and suggest directions for future research. We hope MuCR can serve as a valuable resource and foundational benchmark in multimodal causal reasoning research. The project is available at: https://github.com/Zhiyuan-Li-John/MuCR
Foundation models are first pre-trained on vast unsupervised datasets and then fine-tuned on labeled data. Reinforcement learning, notably from human feedback (RLHF), can further align the network with the intended usage. Yet the imperfections in the proxy reward may hinder the training and lead to suboptimal results; the diversity of objectives in real-world tasks and human opinions exacerbate the issue. This paper proposes embracing the heterogeneity of diverse rewards by following a multi-policy strategy. Rather than focusing on a single a priori reward, we aim for Pareto-optimal generalization across the entire space of preferences. To this end, we propose rewarded soup, first specializing multiple networks independently (one for each proxy reward) and then interpolating their weights linearly. This succeeds empirically because we show that the weights remain linearly connected when fine-tuned on diverse rewards from a shared pre-trained initialization. We demonstrate the effectiveness of our approach for text-to-text (summarization, Q&A, helpful assistant, review), text-image (image captioning, text-to-image generation, visual grounding, VQA), and control (locomotion) tasks. We hope to enhance the alignment of deep models, and how they interact with the world in all its diversity.
The persistent challenge of medical image synthesis posed by the scarcity of annotated data and the need to synthesize `missing modalities' for multi-modal analysis, underscored the imperative development of effective synthesis methods. Recently, the combination of Low-Rank Adaptation (LoRA) with latent diffusion models (LDMs) has emerged as a viable approach for efficiently adapting pre-trained large language models, in the medical field. However, the direct application of LoRA assumes uniform ranking across all linear layers, overlooking the significance of different weight matrices, and leading to sub-optimal outcomes. Prior works on LoRA prioritize the reduction of trainable parameters, and there exists an opportunity to further tailor this adaptation process to the intricate demands of medical image synthesis. In response, we present SeLoRA, a Self-Expanding Low-Rank Adaptation Module, that dynamically expands its ranking across layers during training, strategically placing additional ranks on crucial layers, to allow the model to elevate synthesis quality where it matters most. The proposed method not only enables LDMs to fine-tune on medical data efficiently but also empowers the model to achieve improved image quality with minimal ranking. The code of our SeLoRA method is publicly available on https://anonymous.4open.science/r/SeLoRA-980D .
While deep face recognition models have demonstrated remarkable performance, they often struggle on the inputs from domains beyond their training data. Recent attempts aim to expand the training set by relying on computationally expensive and inherently challenging image-space augmentation of image generation modules. In an orthogonal direction, we present a simple yet effective method to expand the training data by interpolating between instance-level feature statistics across labeled and unlabeled sets. Our method, dubbed Targeted Style Adversary (TSA), is motivated by two observations: (i) the input domain is reflected in feature statistics, and (ii) face recognition model performance is influenced by style information. Shifting towards an unlabeled style implicitly synthesizes challenging training instances. We devise a recognizability metric to constraint our framework to preserve the inherent identity-related information of labeled instances. The efficacy of our method is demonstrated through evaluations on unconstrained benchmarks, outperforming or being on par with its competitors while offering nearly a 70\% improvement in training speed and 40\% less memory consumption.
Current human image customization methods leverage Stable Diffusion (SD) for its rich semantic prior. However, since SD is not specifically designed for human-oriented generation, these methods often require extensive fine-tuning on large-scale datasets, which renders them susceptible to overfitting and hinders their ability to personalize individuals with previously unseen styles. Moreover, these methods extensively focus on single-concept human image synthesis and lack the flexibility to customize individuals using multiple given concepts, thereby impeding their broader practical application. This paper proposes MagicFace, a novel training-free method for multi-concept universal-style human image personalized synthesis. Our core idea is to simulate how humans create images given specific concepts, i.e., first establish a semantic layout considering factors such as concepts' shape and posture, then optimize details by comparing with concepts at the pixel level. To implement this process, we introduce a coarse-to-fine generation pipeline, involving two sequential stages: semantic layout construction and concept feature injection. This is achieved by our Reference-aware Self-Attention (RSA) and Region-grouped Blend Attention (RBA) mechanisms. In the first stage, RSA enables the latent image to query features from all reference concepts simultaneously, extracting the overall semantic understanding to facilitate the initial semantic layout establishment. In the second stage, we employ an attention-based semantic segmentation method to pinpoint the latent generated regions of all concepts at each step. Following this, RBA divides the pixels of the latent image into semantic groups, with each group querying fine-grained features from the corresponding reference concept. Extensive experiments demonstrate the superiority of our MagicFace.
Generative diffusion models, famous for their performance in image generation, are popular in various cross-domain applications. However, their use in the communication community has been mostly limited to auxiliary tasks like data modeling and feature extraction. These models hold greater promise for fundamental problems in network optimization compared to traditional machine learning methods. Discriminative deep learning often falls short due to its single-step input-output mapping and lack of global awareness of the solution space, especially given the complexity of network optimization's objective functions. In contrast, generative diffusion models can consider a broader range of solutions and exhibit stronger generalization by learning parameters that describe the distribution of the underlying solution space, with higher probabilities assigned to better solutions. We propose a new framework Diffusion Model-based Solution Generation (DiffSG), which leverages the intrinsic distribution learning capabilities of generative diffusion models to learn high-quality solution distributions based on given inputs. The optimal solution within this distribution is highly probable, allowing it to be effectively reached through repeated sampling. We validate the performance of DiffSG on several typical network optimization problems, including mixed-integer non-linear programming, convex optimization, and hierarchical non-convex optimization. Our results demonstrate that DiffSG outperforms existing baseline methods not only on in-domain inputs but also on out-of-domain inputs. In summary, we demonstrate the potential of generative diffusion models in tackling complex network optimization problems and outline a promising path for their broader application in the communication community. Our code is available at https://github.com/qiyu3816/DiffSG.
Synthetic infrared (IR) scene and target generation is an important computer vision problem as it allows the generation of realistic IR images and targets for training and testing of various applications, such as remote sensing, surveillance, and target recognition. It also helps reduce the cost and risk associated with collecting real-world IR data. This survey paper aims to provide a comprehensive overview of the conventional mathematical modelling-based methods and deep learning-based methods used for generating synthetic IR scenes and targets. The paper discusses the importance of synthetic IR scene and target generation and briefly covers the mathematics of blackbody and grey body radiations, as well as IR image-capturing methods. The potential use cases of synthetic IR scenes and target generation are also described, highlighting the significance of these techniques in various fields. Additionally, the paper explores possible new ways of developing new techniques to enhance the efficiency and effectiveness of synthetic IR scenes and target generation while highlighting the need for further research to advance this field.
With recent generative models facilitating photo-realistic image synthesis, the proliferation of synthetic images has also engendered certain negative impacts on social platforms, thereby raising an urgent imperative to develop effective detectors. Current synthetic image detection (SID) pipelines are primarily dedicated to crafting universal artifact features, accompanied by an oversight about SID training paradigm. In this paper, we re-examine the SID problem and identify two prevalent biases in current training paradigms, i.e., weakened artifact features and overfitted artifact features. Meanwhile, we discover that the imaging mechanism of synthetic images contributes to heightened local correlations among pixels, suggesting that detectors should be equipped with local awareness. In this light, we propose SAFE, a lightweight and effective detector with three simple image transformations. Firstly, for weakened artifact features, we substitute the down-sampling operator with the crop operator in image pre-processing to help circumvent artifact distortion. Secondly, for overfitted artifact features, we include ColorJitter and RandomRotation as additional data augmentations, to help alleviate irrelevant biases from color discrepancies and semantic differences in limited training samples. Thirdly, for local awareness, we propose a patch-based random masking strategy tailored for SID, forcing the detector to focus on local regions at training. Comparative experiments are conducted on an open-world dataset, comprising synthetic images generated by 26 distinct generative models. Our pipeline achieves a new state-of-the-art performance, with remarkable improvements of 4.5% in accuracy and 2.9% in average precision against existing methods. Our code is available at: https://github.com/Ouxiang-Li/SAFE.
Stable Diffusion Models (SDMs) have shown remarkable proficiency in image synthesis. However, their broad application is impeded by their large model sizes and intensive computational requirements, which typically require expensive cloud servers for deployment. On the flip side, while there are many compact models tailored for edge devices that can reduce these demands, they often compromise on semantic integrity and visual quality when compared to full-sized SDMs. To bridge this gap, we introduce Hybrid SD, an innovative, training-free SDMs inference framework designed for edge-cloud collaborative inference. Hybrid SD distributes the early steps of the diffusion process to the large models deployed on cloud servers, enhancing semantic planning. Furthermore, small efficient models deployed on edge devices can be integrated for refining visual details in the later stages. Acknowledging the diversity of edge devices with differing computational and storage capacities, we employ structural pruning to the SDMs U-Net and train a lightweight VAE. Empirical evaluations demonstrate that our compressed models achieve state-of-the-art parameter efficiency (225.8M) on edge devices with competitive image quality. Additionally, Hybrid SD reduces the cloud cost by 66% with edge-cloud collaborative inference.