Pith. sign in

REVIEW 4 major objections 5 minor 3 cited by

Diffusion Autoencoders are Scalable Image Tokenizers

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a single diffusion L2 loss—Flow Matching, an evidence-lower-bound objective—can train scalable image tokenizers, and that these tokenizers match or outperform the standard GAN-LPIPS tokenizer (GLPTo) for both image…

desk verdict DiTo's real contribution is a simpler self-supervised tokenizer with competitive generation FID; the 'better reconstruction' claim rests on a human eval that is statistically a tie. read the letter →

arxiv 2501.18593 v1 pith:N7WHLAN2 submitted 2025-01-30 cs.CV cs.AIcs.LG

Yinbo Chen , Rohit Girdhar , Xiaolong Wang , Sai Saketh Rambhatla , Ishan Misra This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords imagetokenizersdiffusionautoencodersflowmatchinglatentmodelsself-supervisedlearningreconstructiongenerationELBOobjective
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

This paper asks whether image tokenizers—the autoencoders that compress pixels into compact latents for image generation—need the complex multi-loss recipe used by today's best systems, an autoencoder combining L1, LPIPS, and GAN losses. It argues that one self-supervised diffusion L2 loss, Flow Matching, an evidence-lower-bound objective, is enough to train a tokenizer end to end. The resulting diffusion tokenizer (DiTo) matches or improves on that GAN-LPIPS tokenizer (GLPTo) in reconstruction quality and in the quality of image generators trained on its latents. If correct, this removes the need to balance L1, LPIPS, and adversarial losses, and makes tokenizer quality improve predictably with model scale.

What carries the argument

The central object is the diffusion tokenizer: an encoder plus a diffusion decoder trained with a single $L^2$ Flow Matching loss, which the paper justifies as an ELBO (evidence lower bound) maximization objective. The decoder is a convolutional UNet with timestep conditioning that takes the upsampled latent $z$ as a condition and denoises $x_t$ toward $x$; at inference it samples the reconstruction with an Euler ODE solver. Two design choices carry the argument: using an ELBO-form objective (Flow Matching or v-prediction with cosine schedule) rather than non-ELBO alternatives such as $\epsilon$-prediction or EDM, which cause color shifts in jointly trained latents; and noise synchronization, which during training noises $z$ at a random time $\tau$ with probability 0.1 and asks the decoder to denoise pixel noises from $t \in [\tau,1]$, aligning the latent-space diffusion path with the pixel-space path. DiTo also replaces the KL regularizer on $z$ with LayerNorm.

What would settle it

A direct test: collect more human preferences between DiTo-XL and GLPTo-XL at both 256px and 512px; if DiTo no longer wins at either resolution, or if a text-reading metric on symbol-heavy crops shows no advantage, the reconstruction claim fails. For the generation claim, retrain a DiT-XL/2 on DiTo-XL latents with and without noise synchronization over multiple seeds and check whether the gFID 6.29 versus 7.49 improvement reproduces within noise.

Watch

Extended reading notes

Core claim

DiTo is a diffusion autoencoder trained jointly from scratch: an encoder maps an image $x$ to a compact latent $z$, and a decoder models $p(x|z)$ using a Flow Matching objective, which the paper connects to ELBO theory. The discovery is that this single L2 loss learns latents that are competitive with or better than the GAN-LPIPS recipe. At the XL scale, DiTo's reconstruction FID at 5K samples is 7.95 versus GLPTo-XL's 4.14, yet human side-by-side comparisons favor DiTo 52.44% when ties are counted as half, and DiTo-XL has the best PSNR and SSIM among the compared tokenizers; adding the LPIPS loss already used in GLPTo gives DiTo-XL the best rFID (3.53). On generation, a DiT-XL/2 trained on DiTo latents with noise synchronization reaches gFID 6.29 versus 7.49 for the same generator trained on GLPTo-XL latents. The paper also reports that DiTo's reconstruction quality keeps improving as the decoder grows, with no loss-weight rebalancing needed.

Load-bearing premise

The load-bearing premise is that side-by-side human preference is the right test of reconstruction quality when it conflicts with reconstruction FID, because DiTo-XL's rFID@5K (7.95) is nearly double GLPTo-XL's (4.14) while the paper's better-reconstruction claim rests on a 52.44% preference rate.

Editorial extensions

If this is right

  • Tokenizer training can use a single self-supervised diffusion L2 loss, removing the need to tune weights among L1, LPIPS, and GAN losses.
  • Scaling up the diffusion decoder improves reconstruction faithfulness and image-generation quality without rebalancing loss hyperparameters.
  • Latent diffusion models trained on DiTo latents with noise synchronization outperform those trained on GLPTo-XL latents (gFID 6.29 versus 7.49).
  • Using ELBO-form diffusion objectives such as Flow Matching matters for learning the latent representation itself, not just for decoding.
  • DiTo transfers to higher-resolution inputs zero-shot: the rFID gap versus GLPTo narrows from 7.95 versus 4.14 at 256 pixels to 2.32 versus 2.13 at 512 pixels.

Reading between the lines

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

  • If the single-loss recipe holds, the same diffusion objective could plausibly train tokenizers for video, audio, and other continuous signals, since the loss itself is modality-agnostic.
  • The human-preference result suggests that reconstruction FID, computed in a supervised feature space, may be a misleading yardstick for tokenizer quality; preference tests or task-specific metrics could become the primary evidence.
  • Noise synchronization is a transferable idea for any latent diffusion pipeline: aligning how noise destroys latents with how it destroys pixels could improve other autoencoder-plus-diffusion stacks.
  • DiTo's zero-shot resolution generalization hints that fully convolutional diffusion decoders may eventually let one tokenizer serve multiple resolutions without retraining.
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, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes DiTo, a diffusion autoencoder image tokenizer trained with a single Flow Matching diffusion L2 loss, without GAN or perceptual losses. The encoder is the standard LDM encoder, and the decoder is a diffusion UNet conditioned on the latent. The authors argue that a single ELBO-style diffusion objective is sufficient to learn competitive image tokens, and they introduce a noise-synchronization regularization to improve downstream latent diffusion training. They compare DiTo to an LDM-style GAN-LPIPS tokenizer (GLPTo) at three scales (B, L, XL) for reconstruction and for training a DiT latent generator, reporting competitive or better rFID/gFID and a human-preference study. The paper includes theory grounding via the Kingma-Gao ELBO result, ablations of training objectives, LayerNorm versus KL regularization, and a zero-shot higher-resolution evaluation.

Significance. If the central claim holds, a simple self-supervised diffusion objective can replace the complex GAN-LPIPS recipe used by most current latent image/video models, simplifying tokenizer training and removing dependence on pretrained supervised features. The paper is well grounded theoretically, releases code, and evaluates at nontrivial scales. Its strengths include the clean architecture, the connection to ELBO theory, the LayerNorm analysis, and the honest reporting of rFID gaps. However, the empirical support for the headline 'better reconstruction' claim is fragile: the only direct evidence in the self-supervised setting is a human-preference result at XL that is statistically indistinguishable from a tie and conflicts with the reported rFID and LPIPS/IS numbers. The generation results are promising but are reported without repeated runs. The core idea is valuable, but the claims need to be tightened and supported by additional statistical evidence.

major comments (4)
  1. [Section 5.1, Figure 4 and Table 4] The claim that DiTo-XL achieves better reconstruction than GLPTo-XL is based on a human-preference rate of 52.44% when ties are counted as half, from 900 MTurk comparisons. With N=900, the standard error of this proportion is about 1.67 percentage points, so the 95% confidence interval is approximately [49.2%, 55.7%] and includes 50%; the result is statistically indistinguishable from a tie. This matters because the authors explicitly argue that FID is not aligned with human judgment, making the human protocol the load-bearing evidence for the better-reconstruction claim. As reported, the protocol cannot carry that claim. I recommend reporting confidence intervals or pairwise significance tests, increasing the number of comparisons, and providing a robustness analysis with respect to how ties are handled. Without this, the claim should be moderated to 'competitive' rather than 'better.'
  2. [Section 5.2, Table 2] The generation results are based on a single training run of each DiT model. The differences in gFID@50K (7.49 for GLPTo-XL, 7.57 for DiTo-XL, 6.29 for DiTo-XL with noise sync) are small relative to typical run-to-run variance for large generative models. Without multiple seeds or at least a measure of variance, the statement that noise synchronization 'outperforms' GLPTo is not established. Please report multiple runs with error bars or clearly label these as single-run comparisons and adjust the wording accordingly.
  3. [Section 4, Noise synchronization] Noise synchronization is presented as a key contribution, but the paper does not ablate its design choices: the probability p=0.1, the sampling range of tau, and the exact conditioning of the decoder on z_tau with t in [tau,1]. It is therefore unclear whether the gFID improvement in Table 2 is attributable to the proposed regularization or to a particular choice of hyperparameters. Please provide an ablation of p and tau, and clarify the notation around 'x_tau...1' and the used loss, as the current description is difficult to follow.
  4. [Section 5.3, Figure 5] The comparison of ELBO versus non-ELBO diffusion objectives (Flow Matching, v-prediction cosine, epsilon-prediction, EDM) is reported only with qualitative examples. The paper claims that non-ELBO objectives lead to suboptimal latent representations, but no quantitative rFID or generation numbers are given for these variants. Please add quantitative results to support the claim that the ELBO-form objective is important for learning useful latents, especially since this is one of the primary design justifications.
minor comments (5)
  1. [Abstract and Conclusion] The phrase 'competitive or better quality' should be qualified in light of the statistical evidence; the current wording overstates the reconstruction result.
  2. [Table 1] The row labeled 'Supervised (Rombach et al., 2022)' is ambiguous; please clarify that this is the original Stable Diffusion tokenizer, presumably evaluated under the same protocol, and note any differences in training data.
  3. [Figure 4] The preference plot would benefit from error bars or confidence intervals; as it stands, the reader cannot assess the uncertainty of the 52.44% point estimate.
  4. [Appendix E, Equations (10)--(13)] The derivation of the sample prediction x-bar_theta is concise but would be easier to follow if the matrix inverse and the use of the L2-optimal network prediction were motivated more explicitly.
  5. [Appendix F, Table 8] Please state whether the 512x512 rFID evaluation is performed on the same validation image set as the 256x256 evaluation, since the zero-shot generalization claim depends on the comparability of the image distributions.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: DiTo's central claim is empirically tested against an external baseline, and its theoretical grounding rests on an external ELBO result rather than on the paper's own conclusions.

full rationale

The paper's derivation chain is not circular in any load-bearing way. DiTo is defined by a single diffusion L2 (flow matching) objective, and its reconstruction and generation performance are measured against GLPTo, a baseline re-implemented by the authors with matched architecture and data. The claimed theoretical justification comes from Kingma & Gao's external ELBO equivalence for diffusion objectives; the paper does not derive that equivalence itself or import a uniqueness result from its own prior work. The proposed components — LayerNorm on the latent and noise synchronization — are presented as design choices and evaluated through ablations (Tables 5, 2), not derived from the target results. The only self-citation of note is Girdhar et al. 2023 in the passage arguing that automated metrics such as FID do not correlate well with human judgment; this citation is accompanied by external references (Podell et al., Borji, Jayasumana et al.) and is used to motivate human evaluation rather than to prove a core equation. The paper's human-preference-based claim that DiTo-XL achieves better reconstruction than GLPTo-XL is statistically fragile (52.44% with ties split, N=900, consistent with chance), but this is an evidence-quality and correctness concern, not a circularity of definition or fitting. No equation in the paper reduces to a fitted constant, and no prediction is equivalent to its training input by construction. Overall, the core contribution is an empirical architecture and training-objective study with independent baseline comparisons, so circularity is minimal.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard diffusion theory, an external ELBO result, and a few hand-chosen hyperparameters. No new physical or mathematical entities are introduced. The main load-bearing assumptions are the ELBO interpretation of flow matching and the choice of human preference over FID for reconstruction quality.

free parameters (2)
  • Noise synchronization probability p = 0.1
    Chosen by hand in Section 4 to control how often the latent is noised for the synchronization regularizer. Not fitted to data, but a tuneable design choice.
  • LPIPS loss weight (DiTo+LPIPS variant) = 0.5
    Used only in the optional supervised variant in Appendix E. This weight is a hyperparameter, not derived, and the variant is outside the paper's main self-supervised claim.
assumptions (4)
  • standard math Kingma and Gao (2024): flow matching and v-prediction with cosine schedule are ELBO objectives with data augmentation.
    Invoked in Section 4 'Theoretical justification' to claim DiTo maximizes the ELBO of the input image. The paper relies on this external result rather than proving it.
  • domain assumption The diffusion decoder can learn the conditional distribution p(x|z) using the flow matching objective with the given noise schedule and v-prediction.
    Assumed in Section 4, Equation (9). This is a standard diffusion modeling assumption and is not derived in the paper.
  • domain assumption Human preference is the appropriate arbiter of reconstruction quality when it conflicts with rFID.
    Used in Section 5.1 to claim DiTo-XL reconstructs better than GLPTo-XL despite a worse rFID. The paper argues FID is not aligned with human judgment, but this hierarchy is an assumption.
  • domain assumption LayerNorm on the latent prevents collapse and is compatible with downstream latent diffusion training.
    Motivated in Section 4.1 and Appendix B by a comparison to KL loss. There is no theoretical guarantee, only an empirical motivation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diffusion Autoencoders are Scalable Image Tokenizers." pith.science (2026). https://pith.science/paper/N7WHLAN2

@misc{pith2026250118593,
  author       = {Pith},
  title        = {Pith review of: Diffusion Autoencoders are Scalable Image Tokenizers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N7WHLAN2}},
  note         = {Machine review of arXiv:2501.18593}
}
read the original abstract

Tokenizing images into compact visual representations is a key step in learning efficient and high-quality image generative models. We present a simple diffusion tokenizer (DiTo) that learns compact visual representations for image generation models. Our key insight is that a single learning objective, diffusion L2 loss, can be used for training scalable image tokenizers. Since diffusion is already widely used for image generation, our insight greatly simplifies training such tokenizers. In contrast, current state-of-the-art tokenizers rely on an empirically found combination of heuristics and losses, thus requiring a complex training recipe that relies on non-trivially balancing different losses and pretrained supervised models. We show design decisions, along with theoretical grounding, that enable us to scale DiTo for learning competitive image representations. Our results show that DiTo is a simpler, scalable, and self-supervised alternative to the current state-of-the-art image tokenizer which is supervised. DiTo achieves competitive or better quality than state-of-the-art in image reconstruction and downstream image generation tasks.

Figures

Figures reproduced from arXiv: 2501.18593 by the authors.

Figure 1
Figure 1. Diffusion tokenizer (DiTo) is a diffusion autoencoder with an ELBO objective (e.g., Flow Matching). The input image x is passed into the encoder E to obtain the latent representation, i.e., ‘tokens’ z, a decoder D then learns the distribution p(x|z) with the diffusion objective. E and D are jointly trained from scratch. In contrast, prior work (a) relies on a combination of losses, heuristics, and pretrained models … view at source ↗
Figure 2
Figure 2. Comparison of GAN-LPIPS tokenizer (GLPTo) and diffusion tokenizer (DiTo). [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Scalability of diffusion tokenizers. When increasing the number of trainable parameters in the diffusion decoder from DiTo-B, DiTo-L, to DiTo-XL in the joint training, we observe that the image reconstruction quality keeps improving for structures and textures. Both the visual quality and reconstruction faithfulness are improved when scaling up the diffusion tokenizer. observe to be stable, while it is typically hig… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Comparison for human preference of image recon [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Comparison of training objectives in diffusion tokenizers. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Effectiveness of the latent representation [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Training loss curves of DiTo at different scales. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Zero-shot generalization to tokenizing images at higher resolution. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Number of decoding steps vs. image reconstruction quality. We vary the number of steps in DiTo’s diffusion decoder used for image reconstruction. We use the simple Euler ODE solver and observe that 20 to 50 steps are generally sufficient for good reconstruction quality…
Figure 10
Figure 10. Figure 10: Additional qualitative comparison of tokenizers (at 256 pixel resolution). 17 [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Additional qualitative comparison of tokenizers (at 256 pixel resolution). 18 [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. D-AR: Diffusion via Autoregressive Models

    cs.CV 2025-05 conditional novelty 7.0 of 10

    D-AR recasts pixel-space diffusion as vanilla autoregressive next-token prediction using a diffusion-ordered discrete tokenizer, reaching 2.09 FID on ImageNet 256x256 with a 775M Llama backbone.

  2. SSDD: Single-Step Diffusion Decoder for Efficient Image Tokenization

    cs.CV 2025-10 conditional novelty 6.0 of 10

    A GAN-free, single-step distilled diffusion decoder that reconstructs images from latent codes with better perceptual FID than KL-VAE at higher throughput.

  3. 3D Shape Tokenization via Latent Flow Matching

    cs.CV 2024-12 conditional novelty 6.0 of 10

    Shape Tokens, a compact continuous 3D latent learned by fitting each shape's surface density with flow matching, match specialized baselines across reconstruction, CLIP, generation, and ray intersection tasks.

Reference graph

Works this paper leans on

68 extracted references · 25 canonical work pages · cited by 3 Pith papers

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Albergo, M. S. and Vanden-Eijnden, E. Building normalizing flows with stochastic interpolants. arXiv preprint arXiv:2209.15571, 2022

    arXiv 2022

  3. [3]

    L., Kiros, J

    Ba, J. L., Kiros, J. R., and Hinton, G. E. Layer normalization, 2016. URL https://arxiv.org/abs/1607.06450

    arXiv 2016

  4. [4]

    BEiT : B ert pre-training of image transformers

    Bao, H., Dong, L., and Wei, F. BEiT : B ert pre-training of image transformers. In ICLR, 2022

    work page 2022

  5. [5]

    Improving image generation with better captions

    Betker, J., Goh, G., Jing, L., Brooks, T., Wang, J., Li, L., Ouyang, L., Zhuang, J., Lee, J., Guo, Y., et al. Improving image generation with better captions. Computer Science. https://cdn. openai. com/papers/dall-e-3. pdf, 2 0 (3): 0 8, 2023

    2023

  6. [6]

    Birodkar, V., Barcik, G., Lyon, J., Ioffe, S., Minnen, D., and Dillon, J. V. Sample what you cant compress, 2024. URL https://arxiv.org/abs/2409.02529

    arXiv 2024

  7. [7]

    W., Fidler, S., and Kreis, K

    Blattmann, A., Rombach, R., Ling, H., Dockhorn, T., Kim, S. W., Fidler, S., and Kreis, K. Align your latents: High-resolution video synthesis with latent diffusion models. In CVPR, 2023

    work page 2023

  8. [8]

    Pros and cons of gan evaluation measures

    Borji, A. Pros and cons of gan evaluation measures. Computer vision and image understanding, 179: 0 41--65, 2019

    work page 2019

Show all 68 references
  1. [9]

    Pros and cons of gan evaluation measures: New developments

    Borji, A. Pros and cons of gan evaluation measures: New developments. Computer Vision and Image Understanding, 215: 0 103329, 2022

    2022

  2. [10]

    Emerging properties in self-supervised vision transformers

    Caron, M., Touvron, H., Misra, I., J\'egou, H., Mairal, J., Bojanowski, P., and Joulin, A. Emerging properties in self-supervised vision transformers. In ICCV, 2021

    2021

  3. [11]

    A simple framework for contrastive learning of visual representations

    Chen, T., Kornblith, S., Norouzi, M., and Hinton, G. A simple framework for contrastive learning of visual representations. In ICML, 2020

    2020

  4. [12]

    Image neural field diffusion models

    Chen, Y., Wang, O., Zhang, R., Shechtman, E., Wang, X., and Gharbi, M. Image neural field diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 8007--8017, 2024

    2024

  5. [13]

    Emu: Enhancing image generation models using photogenic needles in a haystack

    Dai, X., Hou, J., Ma, C.-Y., Tsai, S., Wang, J., Wang, R., Zhang, P., Vandenhende, S., Wang, X., Dubey, A., et al. Emu: Enhancing image generation models using photogenic needles in a haystack. arXiv preprint arXiv:2309.15807, 2023

    2023 arXiv

  6. [14]

    Imagenet: A large-scale hierarchical image database

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp.\ 248--255. Ieee, 2009

    2009

  7. [15]

    and Nichol, A

    Dhariwal, P. and Nichol, A. Diffusion models beat gans on image synthesis. Advances in neural information processing systems, 34: 0 8780--8794, 2021

    2021

  8. [16]

    Adversarial feature learning

    Donahue, J., Krahenb \"u hl, P., and Darrell, T. Adversarial feature learning. In ICLR, 2016

    2016

  9. [17]

    Taming transformers for high-resolution image synthesis

    Esser, P., Rombach, R., and Ommer, B. Taming transformers for high-resolution image synthesis. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 12873--12883, 2021

    2021

  10. [18]

    S., Shah, A., Yin, X., Parikh, D., and Misra, I

    Girdhar, R., Singh, M., Brown, A., Duval, Q., Azadi, S., Rambhatla, S. S., Shah, A., Yin, X., Parikh, D., and Misra, I. Emu video: Factorizing text-to-video generation by explicit image conditioning. arXiv preprint arXiv:2311.10709, 2023

    2023 arXiv

  11. [19]

    Generative adversarial networks

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial networks. Communications of the ACM, 63 0 (11): 0 139--144, 2020

    2020

  12. [20]

    Bootstrap your own latent-a new approach to self-supervised learning

    Grill, J.-B., Strub, F., Altch \'e , F., Tallec, C., Richemond, P., Buchatskaya, E., Doersch, C., Avila Pires, B., Guo, Z., Gheshlaghi Azar, M., et al. Bootstrap your own latent-a new approach to self-supervised learning. NeurIPS, 2020

    2020

  13. [21]

    Momentum contrast for unsupervised visual representation learning

    He, K., Fan, H., Wu, Y., Xie, S., and Girshick, R. Momentum contrast for unsupervised visual representation learning. In CVPR, 2020

    2020

  14. [22]

    Masked autoencoders are scalable vision learners

    He, K., Chen, X., Xie, S., Li, Y., Doll \'a r, P., and Girshick, R. Masked autoencoders are scalable vision learners. In CVPR, 2022

    2022

  15. [23]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., and Hochreiter, S. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017

    2017

  16. [24]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    2020

  17. [25]

    Rethinking fid: Towards a better evaluation metric for image generation

    Jayasumana, S., Ramalingam, S., Veit, A., Glasner, D., Chakrabarti, A., and Kumar, S. Rethinking fid: Towards a better evaluation metric for image generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 9307--9315, 2024

    2024

  18. [26]

    Elucidating the design space of diffusion-based generative models

    Karras, T., Aittala, M., Aila, T., and Laine, S. Elucidating the design space of diffusion-based generative models. In Oh, A. H., Agarwal, A., Belgrave, D., and Cho, K. (eds.), Advances in Neural Information Processing Systems, 2022. URL https://openreview.net/forum?id=k7FuTOWMOc7

    2022

  19. [27]

    Analyzing and improving the training dynamics of diffusion models

    Karras, T., Aittala, M., Lehtinen, J., Hellsten, J., Aila, T., and Laine, S. Analyzing and improving the training dynamics of diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 24174--24184, 2024

    2024

  20. [28]

    and Gao, R

    Kingma, D. and Gao, R. Understanding diffusion objectives as the elbo with simple data augmentation. Advances in Neural Information Processing Systems, 36, 2024

    2024

  21. [29]

    Photo-realistic single image super-resolution using a generative adversarial network

    Ledig, C., Theis, L., Husz \'a r, F., Caballero, J., Cunningham, A., Acosta, A., Aitken, A., Tejani, A., Totz, J., Wang, Z., et al. Photo-realistic single image super-resolution using a generative adversarial network. In Proceedings of the IEEE conference on computer vision an...

    2017

  22. [30]

    Autoregressive image generation without vector quantization

    Li, T., Tian, Y., Li, H., Deng, M., and He, K. Autoregressive image generation without vector quantization. arXiv preprint arXiv:2406.11838, 2024

    2024 arXiv

  23. [31]

    Lipman, Y., Chen, R. T. Q., Ben-Hamu, H., Nickel, M., and Le, M. Flow matching for generative modeling. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=PqvMRDCJT9t

    2023

  24. [32]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Liu, X., Gong, C., and Liu, Q. Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003, 2022

    2022 arXiv

  25. [33]

    and Hutter, F

    Loshchilov, I. and Hutter, F. Decoupled weight decay regularization. In International Conference on Learning Representations, 2019. URL https://openreview.net/forum?id=Bkg6RiCqY7

    2019

  26. [34]

    and Song, Y

    Lu, C. and Song, Y. Simplifying, stabilizing and scaling continuous-time consistency models. arXiv preprint arXiv:2410.11081, 2024

    2024 arXiv

  27. [35]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

    Lu, C., Zhou, Y., Bao, F., Chen, J., Li, C., and Zhu, J. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps. Advances in Neural Information Processing Systems, 35: 0 5775--5787, 2022

    2022

  28. [36]

    Stacked convolutional auto-encoders for hierarchical feature extraction

    Masci, J., Meier, U., Cires, D., and Schmidhuber, J. Stacked convolutional auto-encoders for hierarchical feature extraction. In ICANN, pp.\ 52--59, 2011

    2011

  29. [37]

    and Maaten, L

    Misra, I. and Maaten, L. v. d. Self-supervised learning of pretext-invariant representations. In CVPR, 2020

    2020

  30. [38]

    Glide: Towards photorealistic image generation and editing with text-guided diffusion models

    Nichol, A., Dhariwal, P., Ramesh, A., Shyam, P., Mishkin, P., McGrew, B., Sutskever, I., and Chen, M. Glide: Towards photorealistic image generation and editing with text-guided diffusion models. arXiv preprint arXiv:2112.10741, 2021

    2021 arXiv

  31. [39]

    Nichol, A. Q. and Dhariwal, P. Improved denoising diffusion probabilistic models. In International conference on machine learning, pp.\ 8162--8171. PMLR, 2021

    2021

  32. [40]

    Oord, A. v. d., Li, Y., and Vinyals, O. Representation learning with contrastive predictive coding. In NeurIPS, 2018

    2018

  33. [41]

    Diffuse VAE : Efficient, controllable and high-fidelity generation from low-dimensional latents

    Pandey, K., Mukherjee, A., Rai, P., and Kumar, A. Diffuse VAE : Efficient, controllable and high-fidelity generation from low-dimensional latents. Transactions on Machine Learning Research, 2022. ISSN 2835-8856. URL https://openreview.net/forum?id=ygoNPRiLxw

    2022

  34. [42]

    and Xie, S

    Peebles, W. and Xie, S. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 4195--4205, 2023

    2023

  35. [43]

    L., Pal, C., and Aubreville, M

    Pernias, P., Rampas, D., Richter, M. L., Pal, C., and Aubreville, M. W\"urstchen: An efficient architecture for large-scale text-to-image diffusion models. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=gU58d5QeGv

    2024

  36. [44]

    SDXL : Improving latent diffusion models for high-resolution image synthesis

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., M \"u ller, J., Penna, J., and Rombach, R. SDXL : Improving latent diffusion models for high-resolution image synthesis. In The Twelfth International Conference on Learning Representations, 2024. URL https://open...

    2024

  37. [45]

    Movie gen: A cast of media foundation models

    Polyak, A., Zohar, A., Brown, A., Tjandra, A., Sinha, A., Lee, A., Vyas, A., Shi, B., Ma, C.-Y., Chuang, C.-Y., et al. Movie gen: A cast of media foundation models. arXiv preprint arXiv:2410.13720, 2024

    2024 arXiv

  38. [46]

    Diffusion autoencoders: Toward a meaningful and decodable representation

    Preechakul, K., Chatthee, N., Wizadwongsa, S., and Suwajanakorn, S. Diffusion autoencoders: Toward a meaningful and decodable representation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 10619--10629, 2022

    2022

  39. [47]

    Hierarchical text-conditional image generation with clip latents

    Ramesh, A., Dhariwal, P., Nichol, A., Chu, C., and Chen, M. Hierarchical text-conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 1 0 (2): 0 3, 2022

    2022 arXiv

  40. [48]

    Unsupervised learning of invariant feature hierarchies with applications to object recognition

    Ranzato, M., Huang, F.-J., Boureau, Y.-L., and LeCun, Y. Unsupervised learning of invariant feature hierarchies with applications to object recognition. In CVPR, 2007

    2007

  41. [49]

    High-resolution image synthesis with latent diffusion models

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 10684--10695, 2022

    2022

  42. [50]

    and Hinton, G

    Salakhutdinov, R. and Hinton, G. Deep B oltzmann machines. In AI-STATS , 2009

    2009

  43. [51]

    and Ho, J

    Salimans, T. and Ho, J. Progressive distillation for fast sampling of diffusion models. In International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id=TIdIXIpzhoI

    2022

  44. [52]

    Multistep distillation of diffusion models via moment matching

    Salimans, T., Mensink, T., Heek, J., and Hoogeboom, E. Multistep distillation of diffusion models via moment matching. arXiv preprint arXiv:2406.04103, 2024

    2024 arXiv

  45. [53]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Sohl-Dickstein, J., Weiss, E., Maheswaranathan, N., and Ganguli, S. Deep unsupervised learning using nonequilibrium thermodynamics. In International conference on machine learning, pp.\ 2256--2265. PMLR, 2015

    2015

  46. [54]

    Denoising diffusion implicit models

    Song, J., Meng, C., and Ermon, S. Denoising diffusion implicit models. In International Conference on Learning Representations, 2021 a . URL https://openreview.net/forum?id=St1giarCHLP

    2021

  47. [55]

    and Dhariwal, P

    Song, Y. and Dhariwal, P. Improved techniques for training consistency models. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=WNzy9bRDvG

    2024

  48. [56]

    and Ermon, S

    Song, Y. and Ermon, S. Generative modeling by estimating gradients of the data distribution. Advances in neural information processing systems, 32, 2019

    2019

  49. [57]

    P., Kumar, A., Ermon, S., and Poole, B

    Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., and Poole, B. Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456, 2020

    2011 arXiv

  50. [58]

    P., Kumar, A., Ermon, S., and Poole, B

    Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., and Poole, B. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021 b . URL https://openreview.net/forum?id=PxTIG12RRHS

    2021

  51. [59]

    Consistency models

    Song, Y., Dhariwal, P., Chen, M., and Sutskever, I. Consistency models. In Proceedings of the 40th International Conference on Machine Learning, ICML'23. JMLR.org, 2023

    2023

  52. [60]

    Extracting and composing robust features with denoising autoencoders

    Vincent, P., Larochelle, H., Bengio, Y., and Manzagol, P.-A. Extracting and composing robust features with denoising autoencoders. In ICML, 2008

    2008

  53. [61]

    Esrgan: Enhanced super-resolution generative adversarial networks

    Wang, X., Yu, K., Wu, S., Gu, J., Liu, Y., Dong, C., Qiao, Y., and Change Loy, C. Esrgan: Enhanced super-resolution generative adversarial networks. In Proceedings of the European conference on computer vision (ECCV) workshops, pp.\ 0--0, 2018

    2018

  54. [62]

    Real-esrgan: Training real-world blind super-resolution with pure synthetic data

    Wang, X., Xie, L., Dong, C., and Shan, Y. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 1905--1914, 2021

    1905

  55. [63]

    P., Hou, T., Wu, Y

    Xie, S., Xiao, Z., Kingma, D. P., Hou, T., Wu, Y. N., Murphy, K. P., Salimans, T., Poole, B., and Gao, R. Em distillation for one-step diffusion models. arXiv preprint arXiv:2405.16852, 2024

    2024 arXiv

  56. [64]

    Yin, T., Gharbi, M., Park, T., Zhang, R., Shechtman, E., Durand, F., and Freeman, W. T. Improved distribution matching distillation for fast image synthesis. In NeurIPS, 2024 a

    2024

  57. [65]

    T., and Park, T

    Yin, T., Gharbi, M., Zhang, R., Shechtman, E., Durand, F., Freeman, W. T., and Park, T. One-step diffusion with distribution matching distillation. In CVPR, 2024 b

    2024

  58. [66]

    Y., Luong, T., Baid, G., Wang, Z., Vasudevan, V., Ku, A., Yang, Y., Ayan, B

    Yu, J., Xu, Y., Koh, J. Y., Luong, T., Baid, G., Wang, Z., Vasudevan, V., Ku, A., Yang, Y., Ayan, B. K., et al. Scaling autoregressive models for content-rich text-to-image generation. arXiv preprint arXiv:2206.10789, 2022

    2022 arXiv

  59. [67]

    A., Shechtman, E., and Wang, O

    Zhang, R., Isola, P., Efros, A. A., Shechtman, E., and Wang, O. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR, 2018

    2018

  60. [68]

    -vae: Denoising as visual decoding, 2024

    Zhao, L., Woo, S., Wan, Z., Li, Y., Zhang, H., Gong, B., Adam, H., Jia, X., and Liu, T. -vae: Denoising as visual decoding, 2024. URL https://arxiv.org/abs/2410.04081

    2024 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.