Pith. sign in

REVIEW 2 major objections 2 minor 154 cited by

Demystifying MMD GANs

T0 review · 2 major / 2 minor · reviewed 2026-05-15 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Gradient estimators for MMD GANs and Wasserstein GANs are unbiased, but finite-sample discriminators bias the generator updates.

desk verdict MMD GANs match WGAN bias properties with fixed critics but pick up generator bias from sample-based discriminator training, plus a usable new Kernel Inception Distance metric. read the letter →

arxiv 1801.01401 v5 pith:2FGXYWYD submitted 2018-01-04 stat.ML cs.LG

Miko{l}aj Bińkowski , Danica J. Sutherland , Michael Arbel , Arthur Gretton This is my paper · ORCID
classification stat.MLcs.LG
keywords MMDGANWassersteingradientbiasunbiasedestimatorsKernelInceptionDistanceintegralprobabilitymetricsgenerativeadversarialnetworkschoice
checked against Cost.FunctionalEquation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper clarifies the bias picture in MMD-based generative adversarial networks and related Wasserstein models. It establishes that the gradient estimators applied during optimization remain unbiased for both the MMD critic and the Wasserstein critic. At the same time, training the discriminator itself on finite samples produces biased gradients with respect to the generator parameters. This distinction matters for practitioners because it explains sources of instability and points toward simpler network choices that still match performance. The work further shows that MMD GANs can adopt smaller critic networks than Wasserstein GANs, yielding faster training, and introduces the Kernel Inception Distance as a convergence diagnostic that can adapt learning rates on the fly.

What carries the argument

The MMD critic whose gradient estimators are shown to be unbiased when the kernel is fixed, together with the finite-sample bias that appears once the discriminator is learned from data.

What would settle it

Train an MMD GAN critic on an effectively infinite data set and verify whether the observed generator gradients exactly match the closed-form unbiased estimator derived in the paper.

Watch

Extended reading notes

Core claim

We show that gradient estimators used in the optimization process for both MMD GANs and Wasserstein GANs are unbiased, but learning a discriminator based on samples leads to biased gradients for the generator parameters. We also discuss the issue of kernel choice for the MMD critic, and characterize the kernel corresponding to the energy distance used for the Cramer GAN critic. Being an integral probability metric, the MMD benefits from training strategies recently developed for Wasserstein GANs. In experiments, the MMD GAN is able to employ a smaller critic network than the Wasserstein GAN, resulting in a simpler and faster-training algorithm with matching performance. We also propose an 2

Load-bearing premise

The theoretical unbiasedness of the critic gradients assumes the kernel is fixed and positive definite, and that any remaining finite-sample bias does not dominate other optimization difficulties.

Editorial extensions

If this is right

  • MMD GANs can use smaller critic networks than Wasserstein GANs while achieving matching performance.
  • Training strategies developed for Wasserstein GANs transfer directly to MMD GANs because both rely on integral probability metrics.
  • The Kernel Inception Distance can serve as a dynamic learning-rate scheduler during GAN training.
  • The kernel corresponding to the energy distance is explicitly characterized, allowing direct comparison between Cramer GAN and MMD GAN critics.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The sample-induced bias identified here may be one concrete mechanism behind the well-known instability of many GAN training runs.
  • Similar unbiasedness proofs could be attempted for other integral probability metric critics, potentially unifying design rules across a wider family of GAN variants.
  • Adaptive use of the Kernel Inception Distance might improve convergence monitoring in non-image generative tasks where FID-style metrics are unavailable.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 2 minor

Summary. The paper investigates MMD GANs and provides a theoretical clarification that gradient estimators for both MMD GANs and Wasserstein GANs are unbiased when the critic is fixed, while learning a discriminator from samples induces bias in the generator gradients. It discusses kernel choice for the MMD critic, characterizes the kernel for the energy distance used in Cramer GANs, and proposes the Kernel Inception Distance (KID) as an improved convergence measure that can be used to adapt learning rates dynamically. Experiments show that MMD GANs achieve matching performance to WGANs using smaller critic networks, resulting in simpler and faster training.

Significance. If the central distinction between population-level unbiasedness (via U-statistics for fixed positive-definite kernels) and finite-sample bias holds, the work offers a useful clarification of gradient issues in integral probability metric GANs, extending prior WGAN results with an independent derivation. The empirical finding that smaller critics suffice and the introduction of KID for practical training provide concrete value for the field.

major comments (2)
  1. [Abstract and theoretical analysis] Abstract and theoretical section: the claim that gradient estimators are unbiased for fixed-critic MMD relies on interchanging gradient and expectation under a fixed positive-definite kernel. When the critic is a neural network, the effective kernel depends on critic parameters; the paper should explicitly state whether critic parameters are held fixed during the generator gradient computation and provide the precise conditions under which the interchange remains valid.
  2. [Experiments] Experiments section: the claim of matching performance with smaller networks is central to the practical contribution, yet no variance across random seeds, multiple runs, or statistical significance tests are reported. This makes it difficult to assess whether the observed equivalence is robust or could be due to training variability.
minor comments (2)
  1. [Abstract] The abstract introduces KID without a one-sentence definition; adding a brief parenthetical description would improve readability.
  2. [Kernel discussion] In the kernel characterization for the energy distance, ensure the final kernel expression is numbered as an equation and the derivation steps are clearly separated from surrounding text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback on our manuscript. We address each major comment below and will incorporate clarifications and additional reporting in the revised version.

read point-by-point responses

  1. Referee: [Abstract and theoretical analysis] Abstract and theoretical section: the claim that gradient estimators are unbiased for fixed-critic MMD relies on interchanging gradient and expectation under a fixed positive-definite kernel. When the critic is a neural network, the effective kernel depends on critic parameters; the paper should explicitly state whether critic parameters are held fixed during the generator gradient computation and provide the precise conditions under which the interchange remains valid.

    Authors: We thank the referee for this observation. In the standard alternating optimization used for MMD GANs (and WGANs), the critic parameters are held fixed during the generator update step; only the generator parameters are optimized while the kernel induced by the current critic remains constant. Under this fixed-kernel regime the interchange of gradient and expectation is justified by the dominated convergence theorem for the bounded continuous functions arising from a positive-definite kernel. We will add an explicit paragraph in the theoretical section stating these conditions and confirming that the critic is frozen during generator gradient computation. revision: yes

  2. Referee: [Experiments] Experiments section: the claim of matching performance with smaller networks is central to the practical contribution, yet no variance across random seeds, multiple runs, or statistical significance tests are reported. This makes it difficult to assess whether the observed equivalence is robust or could be due to training variability.

    Authors: We agree that the absence of variance estimates and statistical tests weakens the empirical claim. Although the reported runs were performed with multiple random seeds and produced qualitatively consistent results, we did not include standard deviations or significance tests in the original manuscript. In the revision we will add error bars computed over at least five independent seeds for the key FID/KID curves and include a brief discussion of statistical significance for the observed performance parity between the smaller MMD critic and the larger WGAN critic. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: independent derivation of bias properties

full rationale

The paper's central claims rest on standard properties of U-statistics for the MMD estimator and the ability to interchange gradient and expectation when the kernel is fixed and positive definite. The distinction between population-level unbiasedness of the gradient estimator and finite-sample bias induced by learning the critic is derived directly from these properties without reducing to fitted parameters, self-definitions, or load-bearing self-citations. Training strategies are borrowed from WGAN literature (non-overlapping authors) but the MMD-specific bias analysis is presented as an independent contribution. No step in the provided derivation chain collapses by construction to its inputs.

Assumptions & free parameters 1 free parameters · 1 assumptions · 1 invented entities

The work relies on standard properties of MMD as an integral probability metric and prior results on Wasserstein GAN training strategies; kernel choice is discussed but treated as a design decision rather than a fitted parameter.

free parameters (1)
  • kernel bandwidth or choice
    Discussed as important for MMD critic performance but no specific fitted values reported in abstract.
assumptions (1)
  • domain assumption MMD is an integral probability metric benefiting from WGAN training strategies
    Invoked to justify using existing WGAN techniques with MMD GANs.
invented entities (1)
  • Kernel Inception Distance
    purpose: Measure of GAN convergence using kernel on Inception features
    Newly proposed metric to dynamically adapt learning rates.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Demystifying MMD GANs." pith.science (2026). https://pith.science/paper/2FGXYWYD

@misc{pith2026180101401,
  author       = {Pith},
  title        = {Pith review of: Demystifying MMD GANs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2FGXYWYD}},
  note         = {Machine review of arXiv:1801.01401}
}
read the original abstract

We investigate the training and performance of generative adversarial networks using the Maximum Mean Discrepancy (MMD) as critic, termed MMD GANs. As our main theoretical contribution, we clarify the situation with bias in GAN loss functions raised by recent work: we show that gradient estimators used in the optimization process for both MMD GANs and Wasserstein GANs are unbiased, but learning a discriminator based on samples leads to biased gradients for the generator parameters. We also discuss the issue of kernel choice for the MMD critic, and characterize the kernel corresponding to the energy distance used for the Cramer GAN critic. Being an integral probability metric, the MMD benefits from training strategies recently developed for Wasserstein GANs. In experiments, the MMD GAN is able to employ a smaller critic network than the Wasserstein GAN, resulting in a simpler and faster-training algorithm with matching performance. We also propose an improved measure of GAN convergence, the Kernel Inception Distance, and show how to use it to dynamically adapt learning rates during GAN training.

Discussion (0). Continue with ORCID to comment.

Forward citations

Showing 60 of 154 Pith papers that cite this

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. See all 154 Pith citations

  1. Disentangling 3D from Large Vision-Language Models for Controlled Portrait Generation

    cs.CV 2025-06 conditional novelty 8.0 of 10

    CLIPortrait disentangles camera and geometry information from CLIP embeddings via 2D canonicalization, then prevents distribution collapse with a Jacobian regularizer, enabling text-guided 3D portrait generation from ...

  2. Bridging Online and Offline Handwriting via Differentiable Physical Rendering

    cs.CV 2026-08 conditional novelty 7.0 of 10

    A six-parameter differentiable brush model plus a diffusion enhancer lets one framework generate handwriting trajectories, rendered images, and polished offline images from text and style references.

  3. Layering Virtual Try-On

    cs.CV 2026-07 conditional novelty 7.0 of 10

    A two-stage diffusion pipeline and new benchmark let virtual try-on add, remove, or swap clothing layers while preserving inner layers, with SOTA results on the new LVTON benchmark and on VITON-HD/DressCode.

  4. Learning a Maximum Entropy Model for Visual Textures using Diffusion

    cs.CV 2026-06 unverdicted novelty 7.0 of 10

    A diffusion-trained maximum entropy model uses 512 learned statistics to synthesize visual textures at quality matching or exceeding prior models that rely on ~177k statistics.

  5. A Unifying Framework for Concept-Based Representational Similarity

    cs.LG 2026-06 unverdicted novelty 7.0 of 10

    A unifying framework decomposes concept alignment into instance-wise and distributional translation and concept consistency, introduces the InterVenchA benchmark, and shows that joint optimization via CoSAE recovers s...

  6. TrioPose: Native Triple-Stream Diffusion Transformers for Pose-Guided Text-to-Image Generation

    cs.CV 2026-06 unverdicted novelty 7.0 of 10

    TrioPose proposes a Triple-Stream Pose-Aware DiT with relational bias masks and spatial loss weighting to achieve SOTA pose-guided text-to-image results on multi-person benchmarks like Human-Art.

  7. Text-to-Image Models Need Less from Text Encoders Than You Think

    cs.CV 2026-06 unverdicted novelty 7.0 of 10

    A bag-of-position-tagged-words embedding guides text-to-image diffusion models as effectively as full contextual text embeddings from standard encoders.

  8. Aligning Few-Step Generative Models by Amortizing Sample-based Variational Inference

    cs.LG 2026-05 unverdicted novelty 7.0 of 10

    FAV aligns few-step generative models by amortizing SVGD updates from reward-tilted sampling into generator parameters via fixed-point regression, requiring only sample access, and shows outperformance on robotics tas...

  9. Distributed Image Compression with Multimodal Side Information at Extremely Low Bitrates

    cs.CV 2026-05 unverdicted novelty 7.0 of 10

    MDIC uses a text-conditioned diffusion decoder and a supervised feature-mask generator on visual side information to achieve SOTA perceptual quality in distributed image compression at extremely low bitrates.

  10. SeamCam: Quantifying Seamless Camouflage via Multi-Cue Visual Detectability

    cs.CV 2026-05 conditional novelty 7.0 of 10

    SeamCam quantifies camouflage by computing one minus the highest IoU recoverable from category-conditioned detection proposals against a ground-truth mask, achieving 78.82% agreement with human judgments.

  11. Active Sampling for Ultra-Low-Bit-Rate Video Compression via Conditional Controlled Diffusion

    cs.CV 2026-05 unverdicted novelty 7.0 of 10

    ActDiff-VC partitions video into segments, transmits adaptive keyframes and budget-aware point trajectories, and reconstructs frames via conditional diffusion, reporting up to 64.6% bitrate reduction at matched NIQE o...

  12. Dress-ED: Instruction-Guided Editing for Virtual Try-On and Try-Off

    cs.CV 2026-03 unverdicted novelty 7.0 of 10

    Dress-ED is the first large-scale benchmark unifying virtual try-on, try-off, and text-guided garment editing with 146k verified samples plus a multimodal diffusion baseline.

  13. Flow-Based Conformal Predictive Distributions

    stat.ML 2026-02 unverdicted novelty 7.0 of 10

    Differentiable nonconformity scores induce flows that sample conformal prediction set boundaries, and mixing flows across levels produces conformal predictive distributions whose quantiles match the sets.

  14. Maximum Mean Discrepancy with Unequal Sample Sizes via Generalized U-Statistics

    stat.ML 2025-12 conditional novelty 7.0 of 10

    With unequal sample sizes, MMD estimators converge under min(nX,nY) scaling, so tests can use all available data instead of discarding surplus samples.

  15. UniMC: Taming Diffusion Transformer for Unified Keypoint-Guided Multi-Class Image Generation

    cs.CV 2025-07 conditional novelty 7.0 of 10

    UniMC uses tokenized instance conditions (class, box, keypoints) and a timestep-aware modulator in a DiT backbone to control multi-class human and animal image generation, trained and evaluated on the new HAIG-2.9M dataset.

  16. MAGIC: Few-Shot Mask-Guided Anomaly Inpainting with Prompt Perturbation, Spatially Adaptive Guidance, and Context Awareness

    cs.CV 2025-07 unverdicted novelty 7.0 of 10

    MAGIC is a few-shot mask-guided anomaly inpainting framework using Gaussian prompt perturbation, spatially adaptive guidance, and context-aware mask alignment to produce high-fidelity, diverse anomalies that outperfor...

  17. Splat and Replace: 3D Reconstruction with Repetitive Elements

    cs.GR 2025-06 conditional novelty 7.0 of 10

    Repetitive objects in 3D scenes are registered into a shared Gaussian representation that propagates well-observed geometry and appearance to poorly observed instances, improving rendered novel views.

  18. Your Demands Deserve More Bits: Referring Semantic Image Compression at Ultra-low Bitrate

    eess.IV 2025-05 conditional novelty 7.0 of 10

    RSIC allocates bits to user-specified image regions via a grounding model and guides a pretrained diffusion decoder with the compressed latent, boosting local fidelity at ultra-low rates.

  19. VTBench: Comprehensive Benchmark Suite Towards Real-World Virtual Try-on Models

    cs.CV 2025-05 conditional novelty 7.0 of 10

    VTBench is a multi-dimensional benchmark with novel unpaired metrics and human preference data for evaluating image-based virtual try-on models, though the human-alignment evidence is incomplete.

  20. Latent Swap Joint Diffusion for 2D Long-Form Latent Generation

    cs.SD 2025-02 conditional novelty 7.0 of 10

    A training-free latent swap method that replaces averaging with binary swapping in joint diffusion, improving long-form audio spectrum and panorama generation.

  21. OccScene: Semantic Occupancy-based Cross-task Mutual Learning for 3D Scene Generation

    cs.CV 2024-12 conditional novelty 7.0 of 10

    A joint diffusion framework trains a Stable Diffusion generator and a semantic occupancy perception model together, so each task improves the other, producing text-conditional RGB-occupancy pairs.

  22. Structure-Detail Decoupled Autoregressive Generation for Fast and High-Fidelity Virtual Try-On

    cs.CV 2026-07 conditional novelty 6.5 of 10

    STAR-VTON decouples latent VAR structure synthesis from pixel-space matching-based detail recovery, yielding faster high-fidelity virtual try-on than diffusion baselines.

  23. DynEval: Holistic Evaluations of T2I Generative Models in the Wild

    cs.CV 2026-07 conditional novelty 6.5 of 10

    DynEval distills a 235B teacher VLM into 2B/4B evaluators via 250K synthetic instruction triplets, yielding higher human correlation than existing T2I metrics while enabling open-set dynamic QA and scene-graph quality checks.

  24. Next-Frame Decoding for Ultra-Low-Bitrate Image Compression with Video Diffusion Priors

    cs.CV 2026-03 conditional novelty 6.5 of 10

    Ultra-low-bitrate image decoding is cast as one-step next-frame prediction from a compact anchor using adapted video diffusion priors, yielding large perceptual bitrate savings versus DiffC.

  25. LayeringDiff: Layered Image Synthesis via Generation, then Disassembly with Generative Knowledge

    cs.CV 2025-01 conditional novelty 6.5 of 10

    LayeringDiff synthesizes layered images by generating a composite with a pretrained diffusion model and then decomposing it into foreground and background layers using small fine-tuned networks.

  26. AdvFD: Boosting Visual Generation via Adversarial Fr'echet Distance Loss

    cs.CV 2026-08 conditional novelty 6.0 of 10

    AdvFD augments static Fréchet-loss post-training with a whitened adversarial representation, mitigating Fréchet hacking and improving one-step ImageNet generation across JiT and pMF backbones.

  27. WearWow: Native 2K Multi-Garment Virtual Try-On via Adaptive Token Packing and Preference Alignment

    cs.CV 2026-07 conditional novelty 6.0 of 10

    WearWow generates native 2K multi-garment virtual try-on images without masks, using token packing plus dual preference rewards to preserve fabric texture.

  28. To Blend In, First Decouple: Rethinking Camouflage Image Generation via Context-Decoupled Representations

    cs.CV 2026-07 conditional novelty 6.0 of 10

    CamoDreamer generates camouflage images by decoupling foreground and background control in a diffusion model, reporting a 15.5-point FID gain over prior state of the art on LAKE-RED.

  29. GRE-Diff: Gaussian Room Embeddings for Structured Layout Diffusion

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Modeling rooms as isotropic Gaussians and using them to initialize and guide diffusion yields controllable, editable polygonal floor plans that beat prior methods on RPLAN similarity and constraint metrics.

  30. ELSA3D: Elastic Semantic Anchoring for Unified 3D Understanding and Generation

    cs.CV 2026-07 unverdicted novelty 6.0 of 10

    ELSA3D introduces elastic semantic anchoring via sparse anchor tokens and a scale-aware octree tokenizer to unify 3D generation and captioning at reduced computational cost.

  31. WarpI2I: Image Warping for Image-to-Image Translation

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    A saliency-guided warp-unwarp method reallocates spatial representation to preserve fine structures in latent diffusion models for image-to-image translation.

  32. SatSplatDiff: Geometry-preserving generative refinement for high-fidelity satellite Gaussian Splatting

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    SatSplatDiff combines depth supervision and shadow-guided generative refinement with prior SatSplat geometry to reduce hallucinations while raising visual quality in satellite 3D reconstruction.

  33. HiFiVe: High-Fidelity Vehicle Generation Leveraging Auto-Regressive 2D Generative Priors

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    HiFiVe is a training-free framework using an auto-regressive texture refinement pipeline with depth-based warping, multi-view fusion, and symmetry to enhance both texture and geometry fidelity in vehicle generation fr...

  34. Lighting-Consistent Object Transfer Across Radiance Fields

    cs.GR 2026-06 unverdicted novelty 6.0 of 10

    Diffusion-based per-view harmonization for lighting-consistent object transfer between 3DGS scenes, using heterogeneous training data and final 3D consolidation.

  35. JanusMesh: Fast and Zero-Shot 3D Visual Illusion Generation via Cross-Space Denoising

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    A training-free two-stage pipeline uses cross-space dual-branch denoising with CLIP-guided voxel alignment and SDF blending for geometry, followed by view-conditioned 2D diffusion texture projection, to produce dual-s...

  36. Training-Free Metrics for Synthetic Object Detection Data: A Proxy for Detector Performance

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    CCDM metrics achieve perfect Spearman correlation of 1.0 with YOLOv8 mAP on VisDrone-DET synthetic sets, outperforming prior synthetic-image metrics.

  37. C3VD-DEFCOL: A Deformable Colonoscopy Dataset with Time-Resolved 3D Ground Truth and Realistic Appearance

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    C3VD-DEFCOL supplies 110 videos from 11 colon meshes with paired realistic RGB appearance and dense time-resolved 3D ground truth under three levels of parameterized peristaltic deformation for benchmarking deformable...

  38. Exploiting Semantic and Pixel Representations for Ultra-Low Bitrate Image Compression

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    SPRDiff is a diffusion model for ultra-low bitrate image compression that fuses features from distortion-oriented, semantic-oriented, and VAE encoders plus a dual-feature reconstruction module to outperform prior meth...

  39. Unlearning in Diffusion Models: A Unified Framework with KL Divergence and Likelihood Constraints

    cs.LG 2026-05 unverdicted novelty 6.0 of 10

    A constrained optimization framework for diffusion model unlearning via KL and likelihood constraints, with duality results and reported better retention-unlearning tradeoffs than weight-based baselines.

  40. TOPOS: High-Fidelity and Efficient Industry-Grade 3D Head Generation

    cs.CV 2026-05 unverdicted novelty 6.0 of 10

    TOPOS creates high-fidelity 3D heads with fixed industry topology from single images via a specialized VAE with Perceiver Resampler and a rectified flow transformer.

  41. GaussianGPT: Towards Autoregressive 3D Gaussian Scene Generation

    cs.CV 2026-03 conditional novelty 6.0 of 10

    A causal transformer with 3D RoPE generates vector-quantized 3D Gaussian latent grids autoregressively, enabling unconditional synthesis, completion, and open-ended outpainting of indoor scenes.

  42. One-to-More: High-Fidelity Training-Free Anomaly Generation with Attention Control

    cs.CV 2026-03 unverdicted novelty 6.0 of 10

    O2MAG generates high-fidelity text-guided anomalies from a single image without training by manipulating self-attention in diffusion models with anomaly masks and dual enhancements.

  43. VTEdit-Bench: A Comprehensive Benchmark for Multi-Reference Image Editing Models in Virtual Try-On

    cs.CV 2026-03 conditional novelty 6.0 of 10

    VTEdit-Bench and VTEdit-QA show top universal multi-reference editors match specialized VTON models on standard tasks and transfer more stably to harder multi-person/multi-cloth settings, yet still fail under complex ...

  44. AutoFFS: Adversarial Deformations for Facial Feminization Surgery Planning

    cs.CV 2026-03 conditional novelty 6.0 of 10

    AutoFFS uses adversarial deformations of skull scans, guided by an ensemble of sex classifiers, to generate counterfactual male/female skull morphologies for surgical planning.

  45. A universal vision transformer for fast calorimeter simulations

    hep-ph 2026-01 conditional novelty 6.0 of 10

    A vision-transformer flow-matching model generates calorimeter showers across regular and irregular detector geometries at millisecond speeds, and pretraining plus fine-tuning cuts training cost by about half.

  46. HyperVQ: Enabling Hyperprior Entropy Modeling for VQ-Based Generative Image Compression

    cs.CV 2025-12 conditional novelty 6.0 of 10

    A hyperprior predicts a Gaussian in codebook space and converts it to index probabilities, enabling content-adaptive entropy coding for VQ image compression.

  47. Weight Space Representation Learning via Neural Field Adaptation

    cs.LG 2025-12 conditional novelty 6.0 of 10

    Multiplicative LoRA weights of pre-trained neural fields form structured, semantically meaningful representations that outperform prior weight-space methods for generation and classification.

  48. RefTon: Reference person shot assist virtual Try-on

    cs.CV 2025-11 unverdicted novelty 6.0 of 10

    RefTon is a flux-based virtual try-on method that uses unpaired reference images of the target garment on different people to guide texture and detail preservation in a streamlined person-to-person pipeline without bo...

  49. One-shot Conditional Sampling: MMD meets Nearest Neighbors

    stat.ML 2025-09 conditional novelty 6.0 of 10

    Conditional distributions can be sampled in one forward pass by training a generator to minimize a nearest-neighbor estimate of expected conditional MMD, with convergence guarantees.

  50. InternScenes: A Large-scale Simulatable Indoor Scene Dataset with Realistic Layouts

    cs.CV 2025-09 unverdicted novelty 6.0 of 10

    InternScenes is a new dataset of approximately 40,000 simulatable indoor scenes that combines real scans, procedural, and designer sources, preserves small objects for realistic layouts, and includes processing for si...

  51. FastFit: Accelerating Multi-Reference Virtual Try-On via Cacheable Diffusion Models

    cs.CV 2025-08 conditional novelty 6.0 of 10

    FastFit uses a cacheable diffusion UNet to compute multi-reference garment features once per generation, enabling about 3.5x faster multi-item virtual try-on with comparable or better fidelity.

  52. CEIDM: A Controlled Entity and Interaction Diffusion Model for Enhanced Text-to-Image Generation

    cs.CV 2025-08 conditional novelty 6.0 of 10

    A training-free extension of InteractDiffusion that uses LLM-mined relations, action feature offsets, and entity masks to improve entity and interaction control in generated images.

  53. CineScale: Free Lunch in High-Resolution Cinematic Visual Generation

    cs.CV 2025-08 conditional novelty 6.0 of 10

    CineScale extends pre-trained diffusion models to 8k image and 4k video generation with mostly tuning-free inference plus a small LoRA adaptation for video.

  54. Delay-constrained re-entry governs large-scale brain seizures and other network pathologies

    q-bio.NC 2025-08 unverdicted novelty 6.0 of 10

    An epilepsy modeling preprint claims delay-constrained re-entry of traveling excitation drives seizures and predicts 184 recorded seizures, but the submitted full text is an unrelated computer vision paper, so the cla...

  55. CatchPhrase: EXPrompt-Guided Encoder Adaptation for Audio-to-Image Generation

    cs.MM 2025-07 conditional novelty 6.0 of 10

    CatchPhrase improves audio-to-image generation by enriching weak class labels with LLM- and audio-caption-based prompts, filtering and retrieving the best prompt per clip, and training a mapping adapter with contrasti...

  56. FW-VTON: Flattening-and-Warping for Person-to-Person Virtual Try-on

    cs.CV 2025-07 conditional novelty 6.0 of 10

    FW-VTON reports state-of-the-art person-to-person virtual try-on results using a flattening, warping, and integration pipeline plus a new P2P-VTON dataset.

  57. Contrastive Conditional-Unconditional Alignment for Long-tailed Diffusion Model

    cs.CV 2025-07 conditional novelty 6.0 of 10

    A two-part training regularizer, an unconditional-only contrastive repulsion plus a large-timestep conditional-unconditional alignment, improves tail-class diversity and fidelity in diffusion models, cutting ImageNet-...

  58. Unsupervised Cardiac Video Translation Via Motion Feature Guided Diffusion Model

    eess.IV 2025-07 conditional novelty 6.0 of 10

    A motion-guided diffusion model synthesizes high-contrast cine cardiac MRI from DENSE cardiac MRI without paired training data.

  59. SemFaceEdit: Semantic Face Editing on Generative Radiance Manifolds

    cs.CV 2025-06 conditional novelty 6.0 of 10

    A 3D-aware GAN editing method that controls geometry and appearance per semantic region (hair, face, garment, background) using semantic-specific latent codes on radiance manifolds.

  60. StableCodec: Taming One-Step Diffusion for Extreme Image Compression

    eess.IV 2025-06 conditional novelty 6.0 of 10

    A one-step diffusion codec that compresses noisy latents at 64x and decodes with a single denoising step, setting state-of-the-art FID, KID, and DISTS at ultra-low bitrates.

See all 154 Pith citations

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages · cited by 154 Pith papers (see all)

  1. [1]

    Towards Principled Methods for Training Generative Adversarial Networks

    M. Arjovsky and L. Bottou. Towards principled methods for training generative adversarial networks. In ICLR, 2017. arXiv:1701.04862

    work page Pith review arXiv 2017

  2. [2]

    Wasserstein GAN

    M. Arjovsky, S. Chintala, and L. Bottou. W asserstein generative adversarial networks. In ICML, 2017. arXiv:1701.07875

    work page Pith review arXiv 2017

  3. [3]

    Do GANs actually learn the distribution? An empirical study

    S. Arora and Y. Zhang. Do GAN s actually learn the distribution? A n empirical study, 2017. arXiv:1706.08224

    work page Pith review arXiv 2017

  4. [4]

    Generalization and Equilibrium in Generative Adversarial Nets (GANs)

    S. Arora, R. Ge, Y. Liang, T. Ma, and Y. Zhang. Generalization and equilibrium in generative adversarial nets ( GAN s). In ICML, 2017. arXiv:1703.00573

    work page Pith review arXiv 2017

  5. [5]

    M. G. Bellemare, I. Danihelka, W. Dabney, S. Mohamed, B. Lakshminarayanan, S. Hoyer, and R. Munos. The C ramer distance as a solution to biased W asserstein gradients, 2017. arXiv:1705.10743

    work page Pith review arXiv 2017

  6. [6]

    Better Mixing via Deep Representations

    Y. Bengio, G. Mesnil, Y. Dauphin, and S. Rifai. Better mixing via deep representations. In ICML, 2013. arXiv:1207.4404

    work page Pith review arXiv 2013

  7. [7]

    BEGAN: Boundary Equilibrium Generative Adversarial Networks

    D. Berthelot, T. Schumm, and L. Metz. BEGAN : Boundary equilibrium generative adversarial networks, 2017. arXiv:1703.10717

    work page Pith review arXiv 2017

  8. [8]

    P. J. Bickel and E. L. Lehmann. Unbiased estimation in convex families. The Annals of Mathematical Statistics, 40 0 (5): 0 1523--1535, 1969

    work page 1969

Show all 62 references
  1. [9]

    Bouchacourt, P

    D. Bouchacourt, P. K. Mudigonda, and S. Nowozin. DISCO nets: DIS similarity CO efficients networks. In NIPS, pp.\ 352--360. 2016

    2016

  2. [10]

    Bounliphone, E

    W. Bounliphone, E. Belilovsky, M. B. Blaschko, I. Antonoglou, and A. Gretton. A test of relative similarity for model selection in generative models. In ICLR, 2016. arXiv:1511.04581

    2016 arXiv

  3. [11]

    Clevert, T

    D.-A. Clevert, T. Unterthiner, and S. Hochreiter. Fast and accurate deep network learning by exponential linear units ( ELU s). In ICLR, 2016. arXiv:1511.07289

    2016 arXiv

  4. [12]

    Danihelka, B

    I. Danihelka, B. Lakshminarayanan, B. Uria, D. Wierstra, and P. Dayan. Comparison of maximum likelihood and GAN -based training of R eal NVP s, 2017. arXiv:1705.05263

    2017 arXiv

  5. [13]

    G. K. Dziugaite, D. M. Roy, and Z. Ghahramani. Training generative neural networks via maximum mean discrepancy optimization. In UAI, 2015. arXiv:1505.03906

    2015 arXiv

  6. [14]

    Fedus, M

    W. Fedus, M. Rosca, B. Lakshminarayanan, A. M. Dai, S. Mohamed, and I. Goodfellow. Many paths to equilibrium: GAN s do not need to decrease a divergence at every step. In ICLR, 2018. arXiv:1710.08446

    2018 arXiv

  7. [15]

    Gneiting and A

    T. Gneiting and A. E. Raftery. Strictly proper scoring rules, prediction, and estimation. JASA, 102 0 (477): 0 359--378, 2007

    2007

  8. [16]

    Goodfellow, J

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio. Generative adversarial nets. In NIPS, 2014. arXiv:1406.2661

    2014 arXiv

  9. [17]

    Gretton, K

    A. Gretton, K. M. Borgwardt, M. J. Rasch, B. Sch \" o lkopf, and A. J. Smola. A kernel two-sample test. JMLR, 13, 2012

    2012

  10. [18]

    Gulrajani, F

    I. Gulrajani, F. Ahmed, M. Arjovsky, V. Dumoulin, and A. Courville. Improved training of W asserstein GAN s. In NIPS, 2017. arXiv:1704.00028

    2017 arXiv

  11. [19]

    Heusel, H

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, G. Klambauer, and S. Hochreiter. GAN s trained by a two time-scale update rule converge to a N ash equilibrium. In NIPS, 2017. arXiv:1706.08500

    2017 arXiv

  12. [20]

    Huang, S

    R. Huang, S. Zhang, T. Li, and R. He. Beyond face rotation: Global and local perception GAN for photorealistic and identity preserving frontal view synthesis. In ICCV, 2017 a . arXiv:1704.04086

    2017 arXiv

  13. [21]

    Huang, Y

    X. Huang, Y. Li, O. Poursaeed, J. Hopcroft, and S. Belongie. Stacked generative adversarial networks. In CVPR, 2017 b . arXiv:1612.04357

    2017 arXiv

  14. [22]

    Y. Jin, K. Zhang, M. Li, Y. Tian, H. Zhu, and Z. Fang. Towards the automatic anime characters creation with generative adversarial networks, 2017. arXiv:1708.05509

    2017 arXiv

  15. [23]

    Kingma and J

    D. Kingma and J. Ba. Adam: A method for stochastic optimization. In ICLR, 2015. arXiv:1412.6980

    2015 arXiv

  16. [24]

    A. Klenke. Probability Theory: A Comprehensive Course. World Publishing Corporation, 2008

    2008

  17. [25]

    Krizhevsky

    A. Krizhevsky. Learning multiple layers of features from tiny images, 2009

    2009

  18. [26]

    LeCun, L

    Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 1998

    1998

  19. [27]

    C. Li, D. Alvarez-Melis, K. Xu, S. Jegelka, and S. Sra. Distributional adversarial networks, 2017 a . arXiv:1706.09549

    2017 arXiv

  20. [28]

    Li, W.-C

    C.-L. Li, W.-C. Chang, Y. Cheng, Y. Yang, and B. P \' o czos. MMD GAN : Towards deeper understanding of moment matching network. In NIPS, 2017 b . arXiv:1705.08584

    2017 arXiv

  21. [29]

    Y. Li, K. Swersky, and R. Zemel. Generative moment matching networks. In ICML, 2015. arXiv:1502.02761

    2015 arXiv

  22. [30]

    L. Liu. On the two-sample statistic approach to generative adversarial networks. Master's thesis, University of Princeton Senior Thesis, April 2017. URL http://arks.princeton.edu/ark:/88435/dsp0179408079v

    2017

  23. [31]

    S. Liu, O. Bousquet, and K. Chaudhuri. Approximation and convergence properties of generative adversarial learning. In NIPS, 2017. arXiv:1705.08991

    2017 arXiv

  24. [32]

    Z. Liu, P. Luo, X. Wang, and X. Tang. Deep learning face attributes in the wild. In ICCV, 2015

    2015

  25. [33]

    Lopez-Paz and M

    D. Lopez-Paz and M. Oquab. Revisiting classifier two-sample tests. In ICLR, 2017. arXiv:1610.06545

    2017 arXiv

  26. [34]

    R. Lyons. Distance covariance in metric spaces. The Annals of Probability, 41 0 (5): 0 3051--3696, 2013

    2013

  27. [35]

    Mityagin

    B. Mityagin. The zero set of a real analytic function, 2015. arXiv:1512.07276

    2015 arXiv

  28. [36]

    Mroueh and T

    Y. Mroueh and T. Sercu. F isher GAN . In NIPS, 2017. arXiv:1705.09675

    2017 arXiv

  29. [37]

    Mroueh, T

    Y. Mroueh, T. Sercu, and V. Goel. McGan : Mean and covariance feature matching GAN . In ICML, 2017. arXiv:1702.08398

    2017 arXiv

  30. [38]

    M \"u ller

    A. M \"u ller. Integral probability metrics and their generating classes of functions. Advances in Applied Probability, 29 0 (2): 0 429--443, 1997

    1997

  31. [39]

    Nowozin, B

    S. Nowozin, B. Cseke, and R. Tomioka. f- GAN : Training generative neural samplers using variational divergence minimization. In NIPS, 2016. arXiv:1606.00709

    2016 arXiv

  32. [40]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine learning in P ython. JMLR, 12: 0 2825--2830, 2011

    2011

  33. [41]

    Piranian

    G. Piranian. The Set of Nondifferentiability of a Continuous Function . The American Mathematical Monthly, 73 0 (4): 0 57--61, 1966

    1966

  34. [42]

    Radford, L

    A. Radford, L. Metz, and S. Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. In ICLR, 2016. arXiv:1511.06434

    2016 arXiv

  35. [43]

    C. E. Rasmussen and C. K. I. Williams. Gaussian Processes for Machine Learning. MIT Press, Cambridge, MA, 2006

    2006

  36. [44]

    Rosenbaum

    S. Rosenbaum. Moments of a truncated bivariate normal distribution. JRSS B, 23: 0 405--408, 1961

    1961

  37. [45]

    Salimans, I

    T. Salimans, I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, and X. Chen. Improved techniques for training GAN s. In NIPS, 2016. arXiv:1606.03498

    2016 arXiv

  38. [46]

    Sejdinovic, B

    D. Sejdinovic, B. K. Sriperumbudur, A. Gretton, and K. Fukumizu. Equivalence of distance-based and RKHS -based statistics in hypothesis testing. The Annals of Stastistics, 41 0 (5): 0 2263--2291, 2013. arXiv:1207.6076

    2013 arXiv

  39. [47]

    B. K. Sriperumbudur, K. Fukumizu, A. Gretton, G. R. G. Lanckriet, and B. Sch \" o lkopf. Kernel choice and classifiability for RKHS embeddings of probability distributions. In NIPS, 2009 a

    2009

  40. [48]

    B. K. Sriperumbudur, K. Fukumizu, A. Gretton, B. Sch \" o lkopf, and G. R. G. Lanckriet. On integral probability metrics, phi-divergences and binary classification, 2009 b . arXiv:0901.2698

    2009 arXiv

  41. [49]

    B. K. Sriperumbudur, A. Gretton, K. Fukumizu, G. R. G. Lanckriet, and B. Sch \"o lkopf. Hilbert space embeddings and metrics on probability measures. JMLR, 11: 0 1517--1561, 2010. arXiv:0907.5309

    2010 arXiv

  42. [50]

    B. K. Sriperumbudur, K. Fukumizu, and G. R. G. Lanckriet. Universality, characteristic kernels and RKHS embedding of measures. JMLR, 12: 0 2389--2410, 2011. arXiv:1003.0887

    2011 arXiv

  43. [51]

    B. K. Sriperumbudur, K. Fukumizu, A. Gretton, B. Sch \" o lkopf, and G. R. G. Lanckriet. On the empirical estimation of integral probability metrics. Electronic Journal of Statistics, 6: 0 1550--1599, 2012

    2012

  44. [52]

    Steinwart and A

    I. Steinwart and A. Christmann. Support Vector Machines. Information Science and Statistics. Springer, 2008

    2008

  45. [53]

    D. J. Sutherland. What are the mean and variance of a 0-censored multivariate normal? Cross Validated answer, 2018. URL https://stats.stackexchange.com/q/326347

    2018

  46. [54]

    D. J. Sutherland, H.-Y. Tung, H. Strathmann, S. De, A. Ramdas, A. Smola, and A. Gretton. Generative models and model criticism via optimized maximum mean discrepancy. In International Conference on Learning Representations, 2017. arXiv:1611.04488

    2017

  47. [55]

    Szegedy, W

    C. Szegedy, W. Zaremba, I. Sutskever, J. Bruna, D. Erhan, I. Goodfellow, and R. Fergus. Intriguing properties of neural networks. In ICLR, 2014. arXiv:1312.6199

    2014 arXiv

  48. [56]

    Szegedy, V

    C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna. Rethinking the I nception architecture for computer vision. In CVPR, 2016. arXiv:1512.00567

    2016 arXiv

  49. [57]

    Sz\' e kely and M

    G. Sz\' e kely and M. Rizzo. Testing for equal distributions in high dimension. InterStat, 5, 2004

    2004

  50. [58]

    Theis, A

    L. Theis, A. van den Oord, and M. Bethge. A note on the evaluation of generative models. In ICLR, 2016. arXiv:1511.01844

    2016 arXiv

  51. [59]

    F. Yu, Y. Zhang, S. Song, A. Seff, and J. Xiao. LSUN : Construction of a large-scale image dataset using deep learning with humans in the loop, 2015. arXiv:1506.03365

    2015 arXiv

  52. [60]

    Zahorski

    Z. Zahorski. Sur l'ensemble des points de non-d \'e rivabilit \'e d'une fonction continue. Bulletin de la Soci \'e t \'e math \'e matique de France , 2: 0 147--178, 1946

    1946

  53. [61]

    Zaremba, A

    W. Zaremba, A. Gretton, and M. B. Blaschko. B-tests: Low variance kernel two-sample tests. In NIPS, 2013. arXiv:1307.1954

    2013 arXiv

  54. [62]

    J.-Y. Zhu, T. Park, P. Isola, and A. A. Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. In ICCV, 2017. arXiv:1703.10593

    2017

Pith tools

Reviewed May 15, 2026 · model on record in the stance chip above.