What's the score?

In recent years, diffusion models (DMs) have drawn significant attention for their success in approximating data distributions, yielding state-of-the-art generative results. Nevertheless, the versatility of these models extends beyond their generative capabilities to encompass various vision applications, such as image inpainting, segmentation, adversarial robustness, among others. This study is dedicated to the investigation of adversarial attacks through the lens of diffusion models. However, our objective does not involve enhancing the adversarial robustness of image classifiers. Instead, our focus lies in utilizing the diffusion model to detect and analyze the anomalies introduced by these attacks on images. To that end, we systematically examine the alignment of the distributions of adversarial examples when subjected to the process of transformation using diffusion models. The efficacy of this approach is assessed across CIFAR-10 and ImageNet datasets, including varying image sizes in the latter. The results demonstrate a notable capacity to discriminate effectively between benign and attacked images, providing compelling evidence that adversarial instances do not align with the learned manifold of the DMs.

To sample from a general target distribution $p_\propto e^{-f_}$ beyond the isoperimetric condition, Huang et al. (2023) proposed to perform sampling through reverse diffusion, giving rise to Diffusion-based Monte Carlo (DMC). Specifically, DMC follows the reverse SDE of a diffusion process that transforms the target distribution to the standard Gaussian, utilizing a non-parametric score estimation. However, the original DMC algorithm encountered high gradient complexity, resulting in an exponential dependency on the error tolerance $\epsilon$ of the obtained samples. In this paper, we demonstrate that the high complexity of DMC originates from its redundant design of score estimation, and proposed a more efficient algorithm, called RS-DMC, based on a novel recursive score estimation method. In particular, we first divide the entire diffusion process into multiple segments and then formulate the score estimation step (at any time step) as a series of interconnected mean estimation and sampling subproblems accordingly, which are correlated in a recursive manner. Importantly, we show that with a proper design of the segment decomposition, all sampling subproblems will only need to tackle a strongly log-concave distribution, which can be very efficient to solve using the Langevin-based samplers with a provably rapid convergence rate. As a result, we prove that the gradient complexity of RS-DMC only has a quasi-polynomial dependency on $\epsilon$, which significantly improves exponential gradient complexity in Huang et al. (2023). Furthermore, under commonly used dissipative conditions, our algorithm is provably much faster than the popular Langevin-based algorithms. Our algorithm design and theoretical framework illuminate a novel direction for addressing sampling problems, which could be of broader applicability in the community.

Despite the growing popularity of diffusion models, gaining a deep understanding of the model class remains somewhat elusive for the uninitiated in non-equilibrium statistical physics. With that in mind, we present what we believe is a more straightforward introduction to diffusion models using directed graphical modelling and variational Bayesian principles, which imposes relatively fewer prerequisites on the average reader. Our exposition constitutes a comprehensive technical review spanning from foundational concepts like deep latent variable models to recent advances in continuous-time diffusion-based modelling, highlighting theoretical connections between model classes along the way. We provide additional mathematical insights that were omitted in the seminal works whenever possible to aid in understanding, while avoiding the introduction of new notation. We envision this article serving as a useful educational supplement for both researchers and practitioners in the area, and we welcome feedback and contributions from the community at https://github.com/biomedia-mira/demystifying-diffusion.

Realistic synthetic tabular data generation encounters significant challenges in preserving privacy, especially when dealing with sensitive information in domains like finance and healthcare. In this paper, we introduce \textit{Federated Tabular Diffusion} (FedTabDiff) for generating high-fidelity mixed-type tabular data without centralized access to the original tabular datasets. Leveraging the strengths of \textit{Denoising Diffusion Probabilistic Models} (DDPMs), our approach addresses the inherent complexities in tabular data, such as mixed attribute types and implicit relationships. More critically, FedTabDiff realizes a decentralized learning scheme that permits multiple entities to collaboratively train a generative model while respecting data privacy and locality. We extend DDPMs into the federated setting for tabular data generation, which includes a synchronous update scheme and weighted averaging for effective model aggregation. Experimental evaluations on real-world financial and medical datasets attest to the framework’s capability to produce synthetic data that maintains high fidelity, utility, privacy, and coverage.

Recent advances in generative diffusion models have enabled the previously unfeasible capability of generating 3D assets from a single input image or a text prompt. In this work, we aim to enhance the quality and functionality of these models for the task of creating controllable, photorealistic human avatars. We achieve this by integrating a 3D morphable model into the state-of-the-art multiview-consistent diffusion approach. We demonstrate that accurate conditioning of a generative pipeline on the articulated 3D model enhances the baseline model performance on the task of novel view synthesis from a single image. More importantly, this integration facilitates a seamless and accurate incorporation of facial expression and body pose control into the generation process. To the best of our knowledge, our proposed framework is the first diffusion model to enable the creation of fully 3D-consistent, animatable, and photorealistic human avatars from a single image of an unseen subject; extensive quantitative and qualitative evaluations demonstrate the advantages of our approach over existing state-of-the-art avatar creation models on both novel view and novel expression synthesis tasks.

Recent years have witnessed remarkable progress in image generation task, where users can create visually astonishing images with high-quality. However, existing text-to-image diffusion models are proficient in generating concrete concepts (dogs) but encounter challenges with more abstract ones (emotions). Several efforts have been made to modify image emotions with color and style adjustments, facing limitations in effectively conveying emotions with fixed image contents. In this work, we introduce Emotional Image Content Generation (EICG), a new task to generate semantic-clear and emotion-faithful images given emotion categories. Specifically, we propose an emotion space and construct a mapping network to align it with the powerful Contrastive Language-Image Pre-training (CLIP) space, providing a concrete interpretation of abstract emotions. Attribute loss and emotion confidence are further proposed to ensure the semantic diversity and emotion fidelity of the generated images. Our method outperforms the state-of-the-art text-to-image approaches both quantitatively and qualitatively, where we derive three custom metrics, i.e., emotion accuracy, semantic clarity and semantic diversity. In addition to generation, our method can help emotion understanding and inspire emotional art design.

Diffusion models have achieved great success in image generation tasks through iterative noise estimation. However, the heavy denoising process and complex neural networks hinder their low-latency applications in real-world scenarios. Quantization can effectively reduce model complexity, and post-training quantization (PTQ), which does not require fine-tuning, is highly promising in accelerating the denoising process. Unfortunately, we find that due to the highly dynamic distribution of activations in different denoising steps, existing PTQ methods for diffusion models suffer from distribution mismatch issues at both calibration sample level and reconstruction output level, which makes the performance far from satisfactory, especially in low-bit cases. In this paper, we propose Enhanced Distribution Alignment for Post-Training Quantization of Diffusion Models (EDA-DM) to address the above issues. Specifically, at the calibration sample level, we select calibration samples based on the density and diversity in the latent space, thus facilitating the alignment of their distribution with the overall samples; and at the reconstruction output level, we propose Fine-grained Block Reconstruction, which can align the outputs of the quantized model and the full-precision model at different network granularity. Extensive experiments demonstrate that EDA-DM outperforms the existing post-training quantization frameworks in both unconditional and conditional generation scenarios. At low-bit precision, the quantized models with our method even outperform the full-precision models on most datasets.

We consider the problem of sampling from an unknown distribution for which only a sufficiently large number of training samples are available. Such settings have recently drawn considerable interest in the context of generative modelling. In this paper, we propose a generative model combining diffusion maps and Langevin dynamics. Diffusion maps are used to approximate the drift term from the available training samples, which is then implemented in a discrete-time Langevin sampler to generate new samples. By setting the kernel bandwidth to match the time step size used in the unadjusted Langevin algorithm, our method effectively circumvents any stability issues typically associated with time-stepping stiff stochastic differential equations. More precisely, we introduce a novel split-step scheme, ensuring that the generated samples remain within the convex hull of the training samples. Our framework can be naturally extended to generate conditional samples. We demonstrate the performance of our proposed scheme through experiments on synthetic datasets with increasing dimensions and on a stochastic subgrid-scale parametrization conditional sampling problem.

Three-dimensional (3D) human pose estimation using a monocular camera has gained increasing attention due to its ease of implementation and the abundance of data available from daily life. However, owing to the inherent depth ambiguity in images, the accuracy of existing monocular camera-based 3D pose estimation methods remains unsatisfactory, and the estimated 3D poses usually include much noise. By observing the histogram of this noise, we find each dimension of the noise follows a certain distribution, which indicates the possibility for a neural network to learn the mapping between noisy poses and ground truth poses. In this work, in order to obtain more accurate 3D poses, a Diffusion-based 3D Pose Refiner (D3PRefiner) is proposed to refine the output of any existing 3D pose estimator. We first introduce a conditional multivariate Gaussian distribution to model the distribution of noisy 3D poses, using paired 2D poses and noisy 3D poses as conditions to achieve greater accuracy. Additionally, we leverage the architecture of current diffusion models to convert the distribution of noisy 3D poses into ground truth 3D poses. To evaluate the effectiveness of the proposed method, two state-of-the-art sequence-to-sequence 3D pose estimators are used as basic 3D pose estimation models, and the proposed method is evaluated on different types of 2D poses and different lengths of the input sequence. Experimental results demonstrate the proposed architecture can significantly improve the performance of current sequence-to-sequence 3D pose estimators, with a reduction of at least 10.3% in the mean per joint position error (MPJPE) and at least 11.0% in the Procrustes MPJPE (P-MPJPE).

Diffusion models have become the go-to method for large-scale generative models in real-world applications. These applications often involve data distributions confined within bounded domains, typically requiring ad-hoc thresholding techniques for boundary enforcement. Reflected diffusion models (Lou23) aim to enhance generalizability by generating the data distribution through a backward process governed by reflected Brownian motion. However, reflected diffusion models may not easily adapt to diverse domains without the derivation of proper diffeomorphic mappings and do not guarantee optimal transport properties. To overcome these limitations, we introduce the Reflected Schrodinger Bridge algorithm: an entropy-regularized optimal transport approach tailored for generating data within diverse bounded domains. We derive elegant reflected forward-backward stochastic differential equations with Neumann and Robin boundary conditions, extend divergence-based likelihood training to bounded domains, and explore natural connections to entropic optimal transport for the study of approximate linear convergence - a valuable insight for practical training. Our algorithm yields robust generative modeling in diverse domains, and its scalability is demonstrated in real-world constrained generative modeling through standard image benchmarks.

Recently, text-to-image diffusion models become a new paradigm in image processing fields, including content generation, image restoration and image-to-image translation. Given a target prompt, Denoising Diffusion Probabilistic Models (DDPM) are able to generate realistic yet eligible images. With this appealing property, the image translation task has the potential to be free from target image samples for supervision. By using a target text prompt for domain adaption, the diffusion model is able to implement zero-shot image-to-image translation advantageously. However, the sampling and inversion processes of DDPM are stochastic, and thus the inversion process often fail to reconstruct the input content. Specifically, the displacement effect will gradually accumulated during the diffusion and inversion processes, which led to the reconstructed results deviating from the source domain. To make reconstruction explicit, we propose a prompt redescription strategy to realize a mirror effect between the source and reconstructed image in the diffusion model (MirrorDiffusion). More specifically, a prompt redescription mechanism is investigated to align the text prompts with latent code at each time step of the Denoising Diffusion Implicit Models (DDIM) inversion to pursue a structure-preserving reconstruction. With the revised DDIM inversion, MirrorDiffusion is able to realize accurate zero-shot image translation by editing optimized text prompts and latent code. Extensive experiments demonstrate that MirrorDiffusion achieves superior performance over the state-of-the-art methods on zero-shot image translation benchmarks by clear margins and practical model stability.

Event cameras or dynamic vision sensors (DVS) record asynchronous response to brightness changes instead of conventional intensity frames, and feature ultra-high sensitivity at low bandwidth. The new mechanism demonstrates great advantages in challenging scenarios with fast motion and large dynamic range. However, the recorded events might be highly sparse due to either limited hardware bandwidth or extreme photon starvation in harsh environments. To unlock the full potential of event cameras, we propose an inventive event sequence completion approach conforming to the unique characteristics of event data in both the processing stage and the output form. Specifically, we treat event streams as 3D event clouds in the spatiotemporal domain, develop a diffusion-based generative model to generate dense clouds in a coarse-to-fine manner, and recover exact timestamps to maintain the temporal resolution of raw data successfully. To validate the effectiveness of our method comprehensively, we perform extensive experiments on three widely used public datasets with different spatial resolutions, and additionally collect a novel event dataset covering diverse scenarios with highly dynamic motions and under harsh illumination. Besides generating high-quality dense events, our method can benefit downstream applications such as object classification and intensity frame reconstruction.

Diffusion models have shown their effectiveness in generation tasks by well-approximating the underlying probability distribution. However, diffusion models are known to suffer from an amplified inherent bias from the training data in terms of fairness. While the sampling process of diffusion models can be controlled by conditional guidance, previous works have attempted to find empirical guidance to achieve quantitative fairness. To address this limitation, we propose a fairness-aware sampling method called \textit{attribute switching} mechanism for diffusion models. Without additional training, the proposed sampling can obfuscate sensitive attributes in generated data without relying on classifiers. We mathematically prove and experimentally demonstrate the effectiveness of the proposed method on two key aspects: (i) the generation of fair data and (ii) the preservation of the utility of the generated data.

Speckle noise poses a significant challenge in maintaining the quality of synthetic aperture radar (SAR) images, so SAR despeckling techniques have drawn increasing attention. Despite the tremendous advancements of deep learning in fixed-scale SAR image despeckling, these methods still struggle to deal with large-scale SAR images. To address this problem, this paper introduces a novel despeckling approach termed Region Denoising Diffusion Probabilistic Model (R-DDPM) based on generative models. R-DDPM enables versatile despeckling of SAR images across various scales, accomplished within a single training session. Moreover, The artifacts in the fused SAR images can be avoided effectively with the utilization of region-guided inverse sampling. Experiments of our proposed R-DDPM on Sentinel-1 data demonstrates superior performance to existing methods.

We propose a novel Latent Diffusion Transformer, namely Latte, for video generation. Latte first extracts spatio-temporal tokens from input videos and then adopts a series of Transformer blocks to model video distribution in the latent space. In order to model a substantial number of tokens extracted from videos, four efficient variants are introduced from the perspective of decomposing the spatial and temporal dimensions of input videos. To improve the quality of generated videos, we determine the best practices of Latte through rigorous experimental analysis, including video clip patch embedding, model variants, timestep-class information injection, temporal positional embedding, and learning strategies. Our comprehensive evaluation demonstrates that Latte achieves state-of-the-art performance across four standard video generation datasets, i.e., FaceForensics, SkyTimelapse, UCF101, and Taichi-HD. In addition, we extend Latte to text-to-video generation (T2V) task, where Latte achieves comparable results compared to recent T2V models. We strongly believe that Latte provides valuable insights for future research on incorporating Transformers into diffusion models for video generation.

This survey delves into the application of diffusion models in time-series forecasting. Diffusion models are demonstrating state-of-the-art results in various fields of generative AI. The paper includes comprehensive background information on diffusion models, detailing their conditioning methods and reviewing their use in time-series forecasting. The analysis covers 11 specific time-series implementations, the intuition and theory behind them, the effectiveness on different datasets, and a comparison among each other. Key contributions of this work are the thorough exploration of diffusion models’ applications in time-series forecasting and a chronologically ordered overview of these models. Additionally, the paper offers an insightful discussion on the current state-of-the-art in this domain and outlines potential future research directions. This serves as a valuable resource for researchers in AI and time-series analysis, offering a clear view of the latest advancements and future potential of diffusion models.

We propose denoising diffusion variational inference (DDVI), an approximate inference algorithm for latent variable models which relies on diffusion models as expressive variational posteriors. Our method augments variational posteriors with auxiliary latents, which yields an expressive class of models that perform diffusion in latent space by reversing a user-specified noising process. We fit these models by optimizing a novel lower bound on the marginal likelihood inspired by the wake-sleep algorithm. Our method is easy to implement (it fits a regularized extension of the ELBO), is compatible with black-box variational inference, and outperforms alternative classes of approximate posteriors based on normalizing flows or adversarial networks. When applied to deep latent variable models, our method yields the denoising diffusion VAE (DD-VAE) algorithm. We use this algorithm on a motivating task in biology – inferring latent ancestry from human genomes – outperforming strong baselines on the Thousand Genomes dataset.

Stable Diffusion XL (SDXL) has become the best open source text-to-image model (T2I) for its versatility and top-notch image quality. Efficiently addressing the computational demands of SDXL models is crucial for wider reach and applicability. In this work, we introduce two scaled-down variants, Segmind Stable Diffusion (SSD-1B) and Segmind-Vega, with 1.3B and 0.74B parameter UNets, respectively, achieved through progressive removal using layer-level losses focusing on reducing the model size while preserving generative quality. We release these models weights at https://hf.co/Segmind. Our methodology involves the elimination of residual networks and transformer blocks from the U-Net structure of SDXL, resulting in significant reductions in parameters, and latency. Our compact models effectively emulate the original SDXL by capitalizing on transferred knowledge, achieving competitive results against larger multi-billion parameter SDXL. Our work underscores the efficacy of knowledge distillation coupled with layer-level losses in reducing model size while preserving the high-quality generative capabilities of SDXL, thus facilitating more accessible deployment in resource-constrained environments.

We introduce a Cascaded Diffusion Model (Cas-DM) that improves a Denoising Diffusion Probabilistic Model (DDPM) by effectively incorporating additional metric functions in training. Metric functions such as the LPIPS loss have been proven highly effective in consistency models derived from the score matching. However, for the diffusion counterparts, the methodology and efficacy of adding extra metric functions remain unclear. One major challenge is the mismatch between the noise predicted by a DDPM at each step and the desired clean image that the metric function works well on. To address this problem, we propose Cas-DM, a network architecture that cascades two network modules to effectively apply metric functions to the diffusion model training. The first module, similar to a standard DDPM, learns to predict the added noise and is unaffected by the metric function. The second cascaded module learns to predict the clean image, thereby facilitating the metric function computation. Experiment results show that the proposed diffusion model backbone enables the effective use of the LPIPS loss, leading to state-of-the-art image quality (FID, sFID, IS) on various established benchmarks.

We introduce a novel sampler called the energy based diffusion generator for generating samples from arbitrary target distributions. The sampling model employs a structure similar to a variational autoencoder, utilizing a decoder to transform latent variables from a simple distribution into random variables approximating the target distribution, and we design an encoder based on the diffusion model. Leveraging the powerful modeling capacity of the diffusion model for complex distributions, we can obtain an accurate variational estimate of the Kullback-Leibler divergence between the distributions of the generated samples and the target. Moreover, we propose a decoder based on generalized Hamiltonian dynamics to further enhance sampling performance. Through empirical evaluation, we demonstrate the effectiveness of our method across various complex distribution functions, showcasing its superiority compared to existing methods.

Diffusion models are a new class of generative models, and have dramatically promoted image generation with unprecedented quality and diversity. Existing diffusion models mainly try to reconstruct input image from a corrupted one with a pixel-wise or feature-wise constraint along spatial axes. However, such point-based reconstruction may fail to make each predicted pixel/feature fully preserve its neighborhood context, impairing diffusion-based image synthesis. As a powerful source of automatic supervisory signal, context has been well studied for learning representations. Inspired by this, we for the first time propose ConPreDiff to improve diffusion-based image synthesis with context prediction. We explicitly reinforce each point to predict its neighborhood context (i.e., multi-stride features/tokens/pixels) with a context decoder at the end of diffusion denoising blocks in training stage, and remove the decoder for inference. In this way, each point can better reconstruct itself by preserving its semantic connections with neighborhood context. This new paradigm of ConPreDiff can generalize to arbitrary discrete and continuous diffusion backbones without introducing extra parameters in sampling procedure. Extensive experiments are conducted on unconditional image generation, text-to-image generation and image inpainting tasks. Our ConPreDiff consistently outperforms previous methods and achieves a new SOTA text-to-image generation results on MS-COCO, with a zero-shot FID score of 6.21.

The diffusion-based Singing Voice Conversion (SVC) methods have achieved remarkable performances, producing natural audios with high similarity to the target timbre. However, the iterative sampling process results in slow inference speed, and acceleration thus becomes crucial. In this paper, we propose CoMoSVC, a consistency model-based SVC method, which aims to achieve both high-quality generation and high-speed sampling. A diffusion-based teacher model is first specially designed for SVC, and a student model is further distilled under self-consistency properties to achieve one-step sampling. Experiments on a single NVIDIA GTX4090 GPU reveal that although CoMoSVC has a significantly faster inference speed than the state-of-the-art (SOTA) diffusion-based SVC system, it still achieves comparable or superior conversion performance based on both subjective and objective metrics. Audio samples and codes are available at https://comosvc.github.io/.

Object detection models represented by YOLO series have been widely used and have achieved great results on the high quality datasets, but not all the working conditions are ideal. To settle down the problem of locating targets on low quality datasets, the existing methods either train a new object detection network, or need a large collection of low-quality datasets to train. However, we propose a framework in this paper and apply it on the YOLO models called DiffYOLO. Specifically, we extract feature maps from the denoising diffusion probabilistic models to enhance the well-trained models, which allows us fine-tune YOLO on high-quality datasets and test on low-quality datasets. The results proved this framework can not only prove the performance on noisy datasets, but also prove the detection results on high-quality test datasets. We will supplement more experiments later (with various datasets and network architectures).

Access to high-quality datasets in the medical industry limits machine learning model performance. To address this issue, we propose a Denoising Diffusion Probabilistic Model (DDPM) combined with a UNet architecture for X-ray image synthesis. Focused on pneumonia medical condition, our methodology employs over 3000 pneumonia X-ray images obtained from Kaggle for training. Results demonstrate the effectiveness of our approach, as the model successfully generated realistic images with low Mean Squared Error (MSE). The synthesized images showed distinct differences from non-pneumonia images, highlighting the model’s ability to capture key features of positive cases. Beyond pneumonia, the applications of this synthesizer extend to various medical conditions, provided an ample dataset is available. The capability to produce high-quality images can potentially enhance machine learning models’ performance, aiding in more accurate and efficient medical diagnoses. This innovative DDPM-based X-ray photo synthesizer presents a promising avenue for addressing the scarcity of positive medical image datasets, paving the way for improved medical image analysis and diagnosis in the healthcare industry.

Diffusion models have emerged as powerful generative tools, rivaling GANs in sample quality and mirroring the likelihood scores of autoregressive models. A subset of these models, exemplified by DDIMs, exhibit an inherent asymmetry: they are trained over $T$ steps but only sample from a subset of $T$ during generation. This selective sampling approach, though optimized for speed, inadvertently misses out on vital information from the unsampled steps, leading to potential compromises in sample quality. To address this issue, we present the S$^{2}$-DMs, which is a new training method by using an innovative $L_{skip}$, meticulously designed to reintegrate the information omitted during the selective sampling phase. The benefits of this approach are manifold: it notably enhances sample quality, is exceptionally simple to implement, requires minimal code modifications, and is flexible enough to be compatible with various sampling algorithms. On the CIFAR10 dataset, models trained using our algorithm showed an improvement of 3.27% to 14.06% over models trained with traditional methods across various sampling algorithms (DDIMs, PNDMs, DEIS) and different numbers of sampling steps (10, 20, …, 1000). On the CELEBA dataset, the improvement ranged from 8.97% to 27.08%. Access to the code and additional resources is provided in the github.