Pith. sign in

REVIEW 3 major objections 4 minor 38 references

Image Classification Using a Diffusion Model as a Pre-Training Model

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a self-supervised diffusion pre-training objective, conditioned on a jointly trained Vision Transformer representation of the input's own latent image, outperforms the contrastive DINOv2 baseline by +6.15% accuracy…

desk verdict Genuine twist on conditioning, but the headline classification gain rests on an unreported timestep and a single unseeded run. read the letter →

arxiv 2505.06890 v1 pith:KWLAVYPD submitted 2025-05-11 cs.LG cs.CVeess.IV

Kosuke Ukita , Ye Xiaolong , Tsuyoshi Okita This is my paper · ORCID
classification cs.LGcs.CVeess.IV
keywords self-supervisedlearningdiffusionmodelslatenttransformerrepresentationconditioningzero-shotclassificationbrainhematomadetectionmedicalimagecontrastive
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 sets out to show that a diffusion model can act as a self-supervised pre-training backbone for image classification, competitive with or better than contrastive learning when labeled data are scarce. It proposes the Representation-Conditioned Latent Diffusion Transformer, where a Vision Transformer encoder maps the clean latent image $z_0$ to a representation $r=f_\phi(z_0)$ that conditions the denoising network at every block, and the encoder is trained jointly with the denoiser by minimizing $\|\epsilon-g_\theta(z_t,r,t)\|^2$. Pre-trained on 10,313 unlabeled brain CT slices and adapted with 1,424 labeled slices, the method classifies hematoma presence through Diffusion Classifier Zero, selecting the class label whose predicted latent image is closest to the input in $\mathbb{E}_{t,\epsilon}\|z_0-z'_0\|^2$. The paper reports accuracy $0.9497$ and F1 $0.8235$, beating DINOv2 by $+6.15\%$ accuracy and $+13.60\%$ F1. If these results hold, diffusion pre-training offers a low-annotation route to high-performance medical image classification.

What carries the argument

The load-bearing object is the Representation-Conditioned Latent Diffusion Transformer, a DiT-style denoiser in the latent space of a fixed VAE, conditioned at two points—after the final decoding layer and at the start of each transformer block—on a representation $r=f_\phi(z_0)$ produced by a Vision Transformer encoder from the noiseless latent image. The representation encoder is trained jointly with the denoiser through the loss $L=\|\epsilon-g_\theta(z_t,r,t)\|^2$, which distinguishes this method from prior work that conditions on features from a frozen pre-trained encoder. Classification is carried out by Diffusion Classifier Zero: for each candidate class label $c$, the tuned denoiser predicts the latent image $z'_0$ from the noised input, and the label minimizing $\mathbb{E}_{t,\epsilon}\|z_0-z'_0(c)\|^2$ is selected. The machinery therefore turns reconstruction fidelity at a controlled timestep into a class decision.

What would settle it

One experiment would settle it: pre-train two representation-conditioned models identically, except that in the control the conditioning encoder receives the latent image of a fixed random image rather than the input's own $z_0$. If the control matches the reported accuracy and F1, the classification gains are not caused by the representation learning the paper claims; if it does not, the conditioning mechanism is doing the work. A second check is to evaluate the reported tuned checkpoints on CT data from a facility excluded from both pre-training and downstream training: if the gains vanish under scanner shift, the acquired features are memorization-like rather than general.

Watch

Extended reading notes

Core claim

The central discovery claimed is that conditioning a latent diffusion transformer on the input's own learned representation makes the denoiser both reconstruct better and transfer better to classification. With representation conditioning, pre-training loss falls to $3.39\times10^{-4}$ compared with $3.13\times10^{-2}$ for the unconditional model, and predicted latent images become close to the original at much earlier timesteps (around $t=800$, versus $t=400$ unconditionally). The paper interprets these as signs that the conditioning signal forces 'deeper training' of the denoising network, producing representations that survive the removal of the conditioning branch and remain useful when the model is re-tuned as a class-conditioned classifier. In the hematoma classification benchmark, that produces accuracy $0.9497$ and F1 $0.8235$, which is $+6.15\%$ accuracy and $+13.60\%$ F1 over DINOv2, and $+6.15\%$ accuracy and $+11.77\%$ F1 over the same architecture without representation conditioning.

Load-bearing premise

The load-bearing premise is that a very low denoising loss during pre-training, achieved by giving the denoiser the image's own representation as a condition, is evidence that the network learned broadly useful features that transfer to classification even though that same representation is discarded during tuning and testing.

Editorial extensions

If this is right

  • If correct, self-supervised diffusion pre-training on unlabeled medical scans can outperform a strong contrastive baseline (DINOv2) on a small downstream labeled set, reducing annotation burden.
  • Representation conditioning improves generation quality as well, with FID dropping to 10.00 from 24.42 (unconditional) and 25.69 (class-conditioned), so one pre-training recipe serves both generation and classification.
  • The large pre-training loss gap ($3.39\times10^{-4}$ vs $3.13\times10^{-2}$) suggests that representation conditioning stabilizes and deepens optimization, which the paper connects to finer and earlier prediction of the latent image.
  • Diffusion Classifier Zero, which classifies by comparing predicted latent images rather than predicted noise, extends zero-shot classification to models pre-trained without labels, avoiding the large labeled dataset required by prior diffusion classifiers.
  • The classification gains are concentrated in F1 and precision ($0.8235$ and $0.9130$, versus $0.6875$ and $0.6111$ for DINOv2), meaning the method reduces false positives in hematoma screening.

Reading between the lines

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

  • Implicit but not established: the classification gain is attributed to 'deeper training' from easier conditioning; a direct test would be to freeze the ViT encoder during pre-training and check whether the downstream gain disappears, which would prove the joint optimization matters.
  • Implicit but not established: the representation $r$ is discarded after pre-training, so an alternative explanation is that the denoiser simply overfits to reconstruction of the pre-training distribution; a domain-shift test with a new scanner or facility would show whether the features generalize.
  • Testable extension: apply the same recipe to other small, imbalanced medical sets such as fundus images or skin lesions, where the paper argues annotation costs are high; the reported $+6.15\%$ accuracy claim is currently a single-dataset result.
  • Testable extension: use $r=f_\phi(z_0)$ directly as a frozen feature extractor for a linear probe; the paper's conditioning premise predicts that this representation should be more useful than the denoising network's hidden states, but this is not reported.
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

3 major / 4 minor

Summary. The paper proposes a self-supervised pre-training framework, the Representation-Conditioned Latent Diffusion Transformer, in which a latent diffusion transformer is conditioned on representations obtained from a ViT encoder that is trained jointly with the denoising network. The model is pre-trained on unlabeled brain CT data, then adapted to a binary hematoma classification task by removing the representation conditioning, adding class-label conditioning, and applying a zero-shot classification rule (Diffusion Classifier Zero) that compares the original latent image with predicted latent images per class. The central claim is that this method outperforms DINOv2 by +6.15% accuracy and +13.60% F1-score on a 179-image test set. The paper also reports generation metrics (FID, sFID, IS, Precision/Recall) and argues that the conditioning mechanism enables lower pre-training loss and higher-quality reconstructions.

Significance. If the claimed classification improvement is robust, the paper would offer a meaningful alternative to contrastive pre-training for small medical datasets, with the added benefit of having a generative model in the same framework. The architecture is clearly described, the use of an unlabeled pre-training corpus is a credible direction, and the paper ships a concrete experimental setting. However, the strength of the evidence is currently insufficient to support the central claim because the classification evaluation omits a key sensitive hyperparameter and lacks any uncertainty quantification, while the generation evaluation is partly circular due to conditioning on the target image's own representation.

major comments (3)
  1. [§3.2 and Table 4] The classification result in Table 4 is not reproducible as reported because the timestep t used in Eq. (4) is not specified, and the paper itself states in §3.2 that classification performance is greatly affected by t and that selecting t takes time. The number of timesteps averaged, the number of noise samples epsilon drawn, and the number of independent pre-training/tuning seeds are also not reported. Without this information, and without confidence intervals or repeated runs, the reported +6.15% accuracy and +13.60% F1 margin over DINOv2 could be a selection artifact rather than a property of the learned representations.
  2. [§5.1 and Eq. (1)] The generation evaluation in Table 3 is confounded by the conditioning mechanism: the representation r is computed as r = f_phi(z0) from the target image's own latent representation (Eq. 1), so the representation-conditioned model is effectively tasked with reconstructing the input rather than generating new samples from the data distribution. The low FID of 10.00 is therefore expected and does not provide independent evidence that the model has learned a meaningful generative model of brain CTs; the comparison against unconditional and class-conditioned models is not a fair head-to-head generation test.
  3. [§5.2 and Figure 8] The comparison of pre-training losses (3.13×10^-2 vs 3.39×10^-4, a difference of about 10^2) is largely by construction: conditioning on the target image's own representation necessarily makes the denoising task easier, since the conditioning channel leaks information about the clean image. The claim that this lower loss indicates 'deeper training' or that it explains the downstream classification gain is not supported by any analysis linking pre-training loss to classifier performance, especially because the representation r is discarded during tuning and testing. The loss difference alone is not evidence that the learned representations are better.
minor comments (4)
  1. [§3.1] There are several typos in this section: 'Represention-Conditioned' should be 'Representation-Conditioned', 'it it combined' should be 'it is combined', 'optimizeing' should be 'optimizing', and 'an arbitary timestep' should be 'an arbitrary timestep'.
  2. [§3.2] The text contains typos such as 'Transfromer' and 'obtaines' in the description of Diffusion Classifier Zero; these should be corrected.
  3. [§4.3] The DINOv2 baseline is underdocumented: the paper does not specify the DINOv2 checkpoint size (ViT-S/B/L), the pre-training setup (number of epochs, optimizer, data augmentations) on the brain CT dataset, or the linear probe / fine-tuning procedure used for the downstream classification. This makes the comparison in Table 4 difficult to interpret.
  4. [§5.1] In Figure 7 the paper generates samples by starting from an intermediate timestep t=100 rather than from complete noise; while this is a reasonable visual comparison, the caption and text should state clearly that these are conditional reconstructions from partially noised latents, not unconditional samples from the learned distribution.

Circularity Check

2 steps flagged · score 6.0 of 10

Generation FID and pre-training loss advantages are by construction because the model is conditioned on the target image's own representation (r=f_phi(z0)); the headline classification gain is independent, but the paper's explanation of it relies on these circular quantities.

  1. self definitional [Section 3.1, Eq. (1); Section 5.1, Table 3 and Figure 7]
    "r =fϕ(z0) (1) ... We believe this result was obtained because providing representations containing information from the original image enables the generation of images closely resembling the original, leading to a closer data distribution."

    Equation (1) defines the conditioning representation r as f_phi(z0), the ViT encoding of the same latent image z0 that the denoiser is asked to reconstruct. In the generation evaluation, the representation-conditioned samples are produced by denoising z_t while conditioning on r from the original z0 (Figure 7 explicitly starts from z_100 of the original), so the 'generated' images are reconstructions of the conditioning input. The FID gap (10.00 vs 24.42) therefore reflects how much information r already carries about the target, not an independent gain in generative quality or in the learned representation. The comparison to unconditional/class-conditioned generation, which do not receive the target image, is not matched.

  2. self definitional [Section 3.1, Eq. (2); Section 5.2, Figure 8]
    "The loss discussed here is the mean squared error between the noise and the predicted noise, as shown in Equation (2). ... Comparing the minimum loss values for each model, the unconditional model achieves 3.13× 10−2, while the representation-conditioned model achieves 3.39× 10−4, indicating a difference of approximately 10^2 scale."

    Equation (2) minimizes L(θ,φ)=||ε−g_θ(z_t,r,t)||^2 with r=f_phi(z0), so the denoising network is given a representation of the exact latent image it must denoise. A much lower loss is the expected by-construction consequence of conditioning on the target's own representation, not an independent measurement of better feature learning. The paper nonetheless uses this 10^2 loss gap to claim that representation-conditioning 'enables deeper training' and to explain the downstream classification improvement, making the explanatory argument circular.

full rationale

The classification result in Table 4 (accuracy .9497 vs .8882 for DINOv2) is not itself circular: during tuning and testing the representation r is removed and the network is conditioned only on class labels, so the +6.15%/+13.60% margin is an independent empirical outcome. However, the paper's supporting evidence for why the method works is substantially circular. The generation FID (Table 3, Figures 6-7) and the ~10^2 lower pre-training loss (Figure 8) are by-construction consequences of conditioning on r=f_phi(z0), the target image's own representation (Eqs. 1-2); the paper explicitly attributes the FID to receiving information from the original image, and uses the loss gap as evidence of 'deeper training.' These quantities cannot independently confirm the value of the learned representation. The paper also flags its own main experimental risk in Section 3.2: 'performance of image classification results is greatly affected, especially by t, and the selection of t takes time,' and Table 4 reports a single accuracy/F1 without specifying t, epsilon samples, repetition seeds, or confidence intervals; this is a reproducibility/selection concern, not a circularity, and is excluded from the score. The only self-citation (Hirano and Okita 2021) is for preprocessing and is not load-bearing. Score is 6 because one or more predictions reduce by construction while the headline classification claim retains independent content.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central empirical result relies on several external modeling choices: the Stable Diffusion VAE latent space, a specific noise schedule, the Diffusion Classifier decision rule, and the transferability of pre-training across scanner facilities. No free parameters are fitted to the test set; however, the unreported timestep t in Diffusion Classifier Zero acts as a tuned hyperparameter.

free parameters (1)
  • Timestep t for Diffusion Classifier Zero = not reported
    Classification accuracy depends strongly on t, as the paper notes by citing Li et al. 2023; the chosen t for each model is not reported, so the headline numbers could reflect tuning on the validation set.
assumptions (4)
  • domain assumption The latent space of a Stable Diffusion VAE trained on ImageNet and LAION provides a suitable representation space for brain CT images.
    Section 3.1 uses z0=E(x) with this VAE for all experiments; no validation of this cross-domain transfer is provided.
  • domain assumption A specific diffusion noise schedule satisfying gamma_t^2 + delta_t^2 = 1 is used, as in DiT.
    The paper imposes only this constraint and references DiT; the exact schedule is not stated, but it affects the diffusion behavior and the results.
  • domain assumption The mean squared error between original and predicted latent image in Diffusion Classifier Zero is a reliable decision rule for binary hematoma classification.
    Section 3.2, Eq. (4); no calibration or comparison to alternative decision rules is provided.
  • domain assumption Pre-training on facilities 2 to 12 transfers to downstream facility 1 despite different scanners and protocols.
    Section 4.1 splits by facility; no domain shift analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Image Classification Using a Diffusion Model as a Pre-Training Model." pith.science (2026). https://pith.science/paper/KWLAVYPD

@misc{pith2026250506890,
  author       = {Pith},
  title        = {Pith review of: Image Classification Using a Diffusion Model as a Pre-Training Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KWLAVYPD}},
  note         = {Machine review of arXiv:2505.06890}
}
read the original abstract

In this paper, we propose a diffusion model that integrates a representation-conditioning mechanism, where the representations derived from a Vision Transformer (ViT) are used to condition the internal process of a Transformer-based diffusion model. This approach enables representation-conditioned data generation, addressing the challenge of requiring large-scale labeled datasets by leveraging self-supervised learning on unlabeled data. We evaluate our method through a zero-shot classification task for hematoma detection in brain imaging. Compared to the strong contrastive learning baseline, DINOv2, our method achieves a notable improvement of +6.15% in accuracy and +13.60% in F1-score, demonstrating its effectiveness in image classification.

Figures

Figures reproduced from arXiv: 2505.06890 by the authors.

Figure 1
Figure 1. illustrates the model structure of DiT. DiT was proposed as a class-conditioned diffusion model, trained by providing class labels as conditions. In this paper, we pro￾pose a new approach for this conditioning, and thus we re￾fer to the generalized model architecture for conditioning as a ”Latent Diffusion Transformer.” In other words, DiT can be regarded as a class-conditioned Latent Diffusion Trans￾former. Since c… view at source ↗
Figure 2
Figure 2. Representation-Conditioned Latent Diffusion [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Self-supervised learning using the Representation-Conditioned Latent Diffusion Transformer [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Diffusion Classifier Zero evaluation by predicted noise, it is difficult to visually distin￾guish the difference between the original noise and the pre￾dicted noise. We propose Diffusion Classifier Zero because we believe it is easier to visually distinguish the differ…
Figure 5
Figure 5. Figure 5: The types of conditions used in this experiment [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Distribution of pre-training data (black) and dis [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 9
Figure 9. Figure 9: zt and z ′ 0 according to time step t around t = 400. For z ′ 0 at t ≥ 600, the image is significantly different from the original image z0and is a low-quality, blurred image. On the other hand, for the representation￾conditioned case, it seems that the original image …
Figure 8
Figure 8. Figure 8: Loss transition during pre-training First, we compare the losses during the pre-training stage [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 6 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    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 word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Beaumont, R. 2021. img2dataset: Easily turn large sets of image urls to an image dataset. https://github.com/rom1504/img2dataset

    work page 2021

  4. [4]

    Bordes, F.; Balestriero, R.; and Vincent, P. 2021. High fidelity visualization of what your self-supervised representation knows about. arXiv preprint arXiv:2112.09164

    arXiv 2021

  5. [5]

    Caron, M.; Touvron, H.; Misra, I.; J \'e gou, H.; Mairal, J.; Bojanowski, P.; and Joulin, A. 2021. Emerging properties in self-supervised vision transformers. In Proceedings of the IEEE/CVF international conference on computer vision, 9650--9660

    2021

  6. [6]

    Chen, T.; Kornblith, S.; Norouzi, M.; and Hinton, G. 2020. A simple framework for contrastive learning of visual representations. In International conference on machine learning, 1597--1607. PMLR

    2020

  7. [7]

    Chen, X.; Liu, Z.; Xie, S.; and He, K. 2024. Deconstructing denoising diffusion models for self-supervised learning. arXiv preprint arXiv:2401.14404

    arXiv 2024

  8. [8]

    Chuang, C.-Y.; Robinson, J.; Yen-Chen, L.; Torralba, A.; and Jegelka, S. 2020. Debiased Contrastive Learning. arXiv:2007.00224

    arXiv 2020

Show all 38 references
  1. [9]

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

    2021

  2. [10]

    Dosovitskiy, A. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

    2020 arXiv

  3. [11]

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

    2021

  4. [12]

    Goodfellow, I.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; and Bengio, Y. 2014. Generative adversarial nets. Advances in neural information processing systems, 27

    2014

  5. [13]

    Gu, S.; Chen, D.; Bao, J.; Wen, F.; Zhang, B.; Chen, D.; Yuan, L.; and Guo, B. 2021. Vector Quantized Diffusion Model for Text-to-Image Synthesis. arXiv preprint arXiv:2111.14822

    2021 arXiv

  6. [14]

    He, K.; Fan, H.; Wu, Y.; Xie, S.; and Girshick, R. 2019. Momentum Contrast for Unsupervised Visual Representation Learning. arXiv preprint arXiv:1911.05722

    2019 arXiv

  7. [15]

    He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770--778

    2016

  8. [16]

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

    2017

  9. [17]

    Hirano, H.; and Okita, T. 2021. Classification of Hematoma: Joint Learning of Semantic Segmentation and Classification. arXiv preprint arXiv:2103.17172

    2021 arXiv

  10. [18]

    Ho, J.; Jain, A.; and Abbeel, P. 2020. Denoising Diffusion Probabilistic Models. arXiv preprint arxiv:2006.11239

    2020 arXiv

  11. [19]

    Kingma, D. P. 2013. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114

    2013 arXiv

  12. [20]

    Krizhevsky, A.; Sutskever, I.; and Hinton, G. E. 2012. ImageNet Classification with Deep Convolutional Neural Networks. In Pereira, F.; Burges, C.; Bottou, L.; and Weinberger, K., eds., Advances in Neural Information Processing Systems, volume 25. Curran Associates, Inc

    2012

  13. [21]

    Kynk \"a \"a nniemi, T.; Karras, T.; Laine, S.; Lehtinen, J.; and Aila, T. 2019. Improved precision and recall metric for assessing generative models. Advances in neural information processing systems, 32

    2019

  14. [22]

    C.; Prabhudesai, M.; Duggal, S.; Brown, E.; and Pathak, D

    Li, A. C.; Prabhudesai, M.; Duggal, S.; Brown, E.; and Pathak, D. 2023. Your diffusion model is secretly a zero-shot classifier. arXiv preprint arXiv:2303.16203

    2023 arXiv

  15. [23]

    Mukhopadhyay, S.; Gwilliam, M.; Agarwal, V.; Padmanabhan, N.; Swaminathan, A.; Hegde, S.; Zhou, T.; and Shrivastava, A. 2023. Diffusion models beat gans on image classification. arXiv preprint arXiv:2307.08702

    2023 arXiv

  16. [24]

    Nash, C.; Menick, J.; Dieleman, S.; and Battaglia, P. W. 2021. Generating images with sparse representations. arXiv preprint arXiv:2103.03841

    2021 arXiv

  17. [25]

    Oquab, M.; Darcet, T.; Moutakanni, T.; Vo, H.; Szafraniec, M.; Khalidov, V.; Fernandez, P.; Haziza, D.; Massa, F.; El-Nouby, A.; et al. 2023. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193

    2023 arXiv

  18. [26]

    Parmar, G.; Zhang, R.; and Zhu, J.-Y. 2022. On aliased resizing and surprising subtleties in gan evaluation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 11410--11420

    2022

  19. [27]

    Paszke, A.; Gross, S.; Massa, F.; Lerer, A.; Bradbury, J.; Chanan, G.; Killeen, T.; Lin, Z.; Gimelshein, N.; Antiga, L.; et al. 2019. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32

    2019

  20. [28]

    Peebles, W.; and Xie, S. 2022. Scalable Diffusion Models with Transformers. arXiv preprint arXiv:2212.09748

    2022 arXiv

  21. [29]

    Robinson, J.; Chuang, C.-Y.; Sra, S.; and Jegelka, S. 2020. Contrastive learning with hard negative samples. arXiv preprint arXiv:2010.04592

    2020 arXiv

  22. [30]

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

    2022

  23. [31]

    Salimans, T.; Goodfellow, I.; Zaremba, W.; Cheung, V.; Radford, A.; and Chen, X. 2016. Improved techniques for training gans. Advances in neural information processing systems, 29

    2016

  24. [32]

    Szegedy, C.; Vanhoucke, V.; Ioffe, S.; Shlens, J.; and Wojna, Z. 2016. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2818--2826

    2016

  25. [33]

    Van der Maaten, L.; and Hinton, G. 2008. Visualizing data using t-SNE. Journal of machine learning research, 9(11)

    2008

  26. [34]

    N.; Kaiser, .; and Polosukhin, I

    Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A. N.; Kaiser, .; and Polosukhin, I. 2017. Attention is all you need. Advances in neural information processing systems, 30

    2017

  27. [35]

    Vincent, P.; Larochelle, H.; Bengio, Y.; and Manzagol, P.-A. 2008. Extracting and composing robust features with denoising autoencoders. In Proceedings of the 25th international conference on Machine learning, 1096--1103

    2008

  28. [36]

    Wei, C.; Mangalam, K.; Huang, P.-Y.; Li, Y.; Fan, H.; Xu, H.; Wang, H.; Xie, C.; Yuille, A.; and Feichtenhofer, C. 2023. Diffusion models as masked autoencoders. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 16284--16294

    2023

  29. [37]

    Xiang, W.; Yang, H.; Huang, D.; and Wang, Y. 2023. Denoising diffusion autoencoders are unified self-supervised learners. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 15802--15812

    2023

  30. [38]

    Yang, X.; and Wang, X. 2023. Diffusion model as representation learner. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 18938--18949

    2023

Pith tools

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