The paper investigates training objectives for denoising-based generative models, focusing on loss weighting and output parameterization such as noise-, clean image-, and velocity-based formulations. It conducts a systematic numerical study across synthetic datasets with controlled geometry and real image data, evaluating denoising accuracy via PSNR and generative quality via FID. The goal is to disentangle how training choices interact with data manifold dimensionality, model architecture, and dataset size, offering practical design insights rather than proposing a new method.
By Anne Gagneux, S\'egol\`ene Martin, R\'emi Gribonval, Mathurin Massias
The paper investigates how current image denoisers handle textures, finding that generic denoising models often over‑smooth fine patterns and that improving texture performance requires fine‑tuning on texture‑specific datasets. However, such fine‑tuning causes the models to lose their ability to denoise the original data, indicating a trade‑off. The authors argue that dedicated, texture‑aware denoisers are necessary for accurate texture reconstruction and for downstream tasks like generative modeling.
By Saptarshi Mandal, Oindrila Haldar, Chirayata Bhattacharya, Chandra Sekhar Seelamantula, Zhiwei Xu, Thrasyvoulos N. Pappas, Gaurav Sharma
The paper proposes a deep learning framework that learns priors for inverse problems by exploiting the relationship between proximal operators and Hamilton–Jacobi partial differential equations. Unlike existing methods that require inverting the prior after training, this approach learns the prior directly, enabling efficient evaluation in a single forward pass. Numerical experiments demonstrate the method’s effectiveness in dimensions up to 64.
By Oluwatosin Akande, Gabriel P. Langlois, Akwum Onwunta
The paper investigates diffusion models trained in a lazy high‑dimensional regime, extending benign overfitting theory to generative settings. By analyzing denoising score matching in a vector‑valued RKHS with an inner‑product kernel, the authors derive exact risk trajectories under gradient flow when the number of samples scales proportionally with dimensionality. These trajectories reveal three distinct phases—spectral generalization, noise‑dominated interpolation, and empirical Bayes memorization—whose interplay shapes the distribution of generated samples.
By Hugo Latourelle-Vigeant, Sinho Chewi, Aram-Alexandre Pooladian, John Sous, Theodor Misiakiewicz
Latent Diffusion-GAN (LD‑GAN) integrates adversarial learning into diffusion models by using the pretrained VAE as an energy‑based discriminator, avoiding extra parameters and preserving the latent priors. It introduces a structural consistency energy to align encoder and decoder features, improving perceptual quality. Experiments show LD‑GAN enhances sample fidelity, sharpness, and diversity across generation tasks while maintaining efficient training dynamics.
The paper investigates how variational autoencoders (VAEs) can be made more learnable in latent diffusion by analyzing diffusability through spectral properties. It introduces the Spectrum Matching Hypothesis, which posits that latents with better diffusability should have a flattened power‑law power spectral density (Encoding Spectrum Matching) and preserve frequency‑to‑frequency semantic correspondence via the decoder (Decoding Spectrum Matching). The authors implement these ideas by matching PSDs between images and latents and applying shared spectral masking, demonstrating improved diffusion generation on CelebA and ImageNet and extending the approach to representation alignment with a DoG‑based method.
By Mang Ning, Mingxiao Li, Le Zhang, Lanmiao Liu, Matthew B. Blaschko, Albert Ali Salah, Itir Onal Ertugrul
arXiv:2606. 04299v1 Announce Type: cross Abstract: We consider the problem of generating images whose internal structure -- defined by the distribution of patches across multiple scales -- matches that of a single reference image.
By Haojun Qiu, Kiriakos N. Kutulakos, David B. Lindell
arXiv:2611.17967v1 Announce Type: new
Abstract: Denoising generative models predict clean data ($x$), noise ($\epsilon$), or velocity ($v$), while training losses can be measured in any of these spac...
By Dongwoo Kim
arXiv:2607. 19332v1 Announce Type: new Abstract: Generative models have undergone many generations of evolution, from VAEs/GANs to diffusion/flow matching.
By Chirag Vashist, Ke Li
The paper introduces Volterra generative models that extend diffusion-based generative models by incorporating Volterra processes, which provide more flexible temporal dependence than traditional Brownian-motion-driven Markov processes. To address the non-Markovian challenges, the authors develop Markovian approximations and Gaussian conditioning, reducing the denoising problem to the original data dimension and enabling connections to variance-exploding and variance-preserving frameworks. The resulting models, trained from scratch or with pretrained denoisers, achieve competitive FID scores on MNIST, conditional CIFAR-10, and ImageNet datasets, demonstrating the practical viability of the approach.
By Bingyan Han
FisherDistill proposes a lightweight refinement for diffusion distillation that replaces learned adversarial discriminators with a closed‑form Fisher linear discriminant. By estimating first‑ and second‑order statistics from frozen teacher representations, it derives a Fisher direction that maximally separates real and generated feature distributions. Experiments across several distillation frameworks show consistent quality improvements and a substantial reduction of the distribution gap compared to adversarially refined baselines.
By Mahesh Bhosale, Yuanhao Zhai, Siwei Lyu, Junsong Yuan, David Doermann, Defang Chen
arXiv:2607. 17673v1 Announce Type: cross Abstract: Contrastive learning is increasingly moving toward settings with three or more modalities instead of image-text pairs.
By Tillmann Rheude, Roland Eils, Benjamin Wild