Pith. sign in

REVIEW 4 major objections 5 minor 53 references

HyperNet Fields: Efficiently Training Hypernetworks without Ground Truth by Learning Weight Trajectories

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A hypernetwork can be trained with no ground-truth weights by learning the full weight trajectory and matching gradient steps.

desk verdict A clever, honest hypernetwork-training idea that avoids precomputed targets; the main claim about gradient-matching being sufficient needs empirical support because the loss is bootstrapped. read the letter →

arxiv 2412.17040 v2 pith:JLDHPFMV submitted 2024-12-22 cs.LG

Eric Hedlin , Munawar Hayat , Fatih Porikli , Kwang Moo Yi , Shweta Mahajan This is my paper · ORCID
classification cs.LG
keywords hypernetworksgradientmatchingweighttrajectoriesneuralfieldspersonalizedimagegeneration3DshapereconstructionDreamBoothoccupancynetworks
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 introduces Hypernetwork Fields, a way to train hypernetworks without precomputing a final optimized weight set for every training sample. The key idea is to give the hypernetwork an extra input, the convergence state t, so it outputs the task network's weights at any point along the training trajectory. Instead of regressing toward known converged weights, the field is supervised solely by matching its own step direction to one gradient-descent step of the task loss computed at the predicted weights. The paper claims this single gradient-matching constraint is sufficient, and demonstrates it on personalized image generation and 3D shape reconstruction, achieving competitive results while cutting training compute to about a quarter of the conventional precompute-based pipeline.

What carries the argument

The central object is the Hypernetwork Field $H_\phi(t,x)$, a hypernetwork augmented with a convergence-state input $t$. It is parameterized as $H_\phi(x,t) = \theta_0 + (t/T) H'_\phi(x,t)$, forcing the field to start at a fixed random initialization $\theta_0$ at $t=0$. The load-bearing mechanism is the gradient-matching loss of Eq. (3), which compares the field's predicted step $H_\phi(x,t+1) - H_\phi(x,t)$ with the true task-gradient step $-\eta \nabla_{\theta_t} L(\theta_t, x)$ computed at the field's own prediction, thereby baking the optimizer dynamics into the hypernetwork without any target weights.

What would settle it

Run the method on a small task where the exact optimizer trajectory can be computed offline, and compare the field's predicted weights at every sampled $t$ against the true trajectory under the same optimizer and learning rate used in Eq. (2); if the predictions diverge beyond a small tolerance while Eq. (3) stays low, or if the endpoint $H_\phi(x,T)$ fails to reach a low task loss, the sufficiency claim is contradicted.

Watch

Extended reading notes

Core claim

The central claim is that a hypernetwork can be trained without any per-sample ground-truth weights, provided it is reparameterized as a field over training time: $H_\phi(t,x)$ predicts the task-network weights at optimization step $t$ for condition $x$. If the field is consistent, its numerical derivative along $t$, $H_\phi(t+1,x) - H_\phi(t,x)$, must equal one gradient step of the task loss evaluated at the predicted weights, $-\eta \nabla_{\theta_t} L(\theta_t, x)$. The paper shows that minimizing the mismatch between these two directions (Eq. 3) is sufficient for the field to learn the entire convergence trajectory, and that at the final timestep $T$ the field's output approximates the converged task network, requiring only a single forward pass at inference.

Load-bearing premise

The whole method assumes that the task network's real training is accurately described as plain gradient descent with a single fixed learning rate, but DreamBooth and occupancy networks are trained with adaptive optimizers and stochastic objectives, so the matched trajectory is only an approximation of the true training path.

Editorial extensions

If this is right

  • Per-sample converged weights no longer need to be precomputed, removing the multi-day preprocessing bottleneck that limits hypernetwork scaling to large datasets.
  • The framework is task-agnostic: swapping the task loss and condition input applies the same gradient-matching training to different base networks, such as diffusion models or occupancy networks.
  • At inference, querying the field once at $t=T$ gives the converged weights, so the method adds no extra compute compared to conventional hypernetwork approaches.
  • Because the field models the whole trajectory, it can also provide partially converged weights, enabling fast fine-tuning or early-stage adaptation from the same learned field.
  • The paper reports that training takes about 1.5 GPU-days versus roughly 50 GPU-days to precompute ground truths for the same scale, a roughly fourfold reduction in total training cost.

Reading between the lines

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

  • A natural extension, not stated in the paper, is to view the gradient-matching objective as amortizing the optimizer itself, potentially allowing the field to learn per-condition learning rates or optimizer hyperparameters in addition to weights.
  • Since the field is supervised at random intermediate timesteps but used at the final timestep, adding an explicit endpoint consistency term at $t=T$ could strengthen the guarantee that the final query actually reaches a good optimum.
  • The method inherits the bias of the assumed task optimizer; testing it with Adam as the task optimizer, rather than plain SGD with fixed $\eta$, would clarify how much of the reported identity preservation comes from the gradient-matching constraint versus the field architecture.
Share X Bluesky LinkedIn Reddit HN

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 HyperNet Fields, a method to train hypernetworks without per-sample precomputed optimized weights. Instead of regressing to final task-network weights, a hypernetwork H_phi(x,t) is trained to predict task-network parameters at any optimization step t, with a shared initialization theta0. The training signal is a self-consistency loss (Eq. 3) that matches the one-step update theta_{t+1} obtained by a gradient-descent step of the task loss to the hypernetwork prediction at t+1. The authors argue that exact satisfaction of this loss implies that H_phi(x,T) equals the result of T gradient steps from theta0. They evaluate the method on DreamBooth personalization (CelebA-HQ, AFHQ) and 3D occupancy reconstruction (ShapeNet, Objaverse), reporting competitive metrics, ablations, and a user study.

Significance. If the central mechanism holds, the paper makes a valuable contribution by removing the costly precompute step of conventional hypernetwork training and by introducing a trajectory-consistency objective that connects to consistency models. The empirical results are promising, particularly the AFHQ results and the user study in Sec. 7. The paper also includes a useful ablation (Sec. 6, Tables 3-4) showing that direct optimization of the task loss via the hypernetwork fails, which strengthens the case for the proposed gradient-matching approach. However, the paper does not provide direct evidence that the trained field satisfies the self-consistency recurrence it is trained against, and it assumes vanilla gradient descent in the derivation while experiments use adaptive optimizers. These gaps temper the strength of the central claim that the gradient-matching constraint alone is sufficient.

major comments (4)
  1. [Sec. 3.2, Eq. (3)] The training objective in Eq. (3) is a self-consistency condition with a moving target: the target theta_{t+1} is computed from the current prediction H_phi(x,t) (Algorithm 1, lines 7-9). The paper does not report the final residual of Eq. (3) nor compare the learned field to the trajectory obtained by directly iterating Eq. (2) from theta0. Figure 7 shows that IoU improves with t for one ShapeNet sample, but IoU is a downstream metric and does not establish that the field satisfies the recurrence in weight space. Without such evidence, the central claim that the gradient-matching constraint alone is sufficient to train a trajectory-accurate hypernetwork is not empirically established. I request a direct weight-space comparison (e.g., L2 or cosine distance between H_phi(x,t) and the iterated theta_t) for a few training and test samples, and a plot of the Eq. (3) loss during training.
  2. [Sec. 3.2, Eq. (2)] The derivation assumes vanilla gradient descent with a fixed learning rate eta, but the task networks in Sec. 4 (DreamBooth and occupancy networks) are normally trained with Adam and, in the case of DreamBooth, a stochastic denoising objective. The paper does not state the value of eta used in Eq. (2), whether it differs from the update rate in Algorithm 1 line 9, or how the simulated vanilla-GD trajectory relates to the actual training dynamics of the task network. This matters because the theoretical fixed-point argument (exact satisfaction of Eq. 3) yields a trajectory defined by Eq. (2), not by Adam or by stochastic gradients. Please clarify the role of eta as a hyperparameter and discuss the implications of the optimizer mismatch for the validity of the trajectory-matching interpretation.
  3. [Sec. 4.2, 3D shape reconstruction] The 3D shape reconstruction experiments are qualitative only. No quantitative comparison is provided against a conventional hypernetwork trained with precomputed weights (e.g., HyperDiffusion) or against per-sample optimized occupancy networks. The Objaverse experiment uses only 128 shapes and is explicitly described as an overfitting demonstration, which does not support the abstract's general claim of 'competitive results' for 3D reconstruction. Please add quantitative IoU or Chamfer-distance comparisons on ShapeNet and, if feasible, on a larger Objaverse subset, or temper the claims accordingly.
  4. [Sec. 4.1, Table 1 and Sec. 6] In Table 1, the Face Rec. score for Ours is 0.325, well below all baselines (0.655, 0.618, 0.623). The authors argue in Sec. 6 that this metric is not critical for stylized prompts, but the table mixes stylized and non-stylized prompts. Since identity preservation is a key claim of the paper, please report Face Rec. separately for non-stylized and stylized prompts, or otherwise justify why the overall score does not weaken the identity-preservation claim.
minor comments (5)
  1. [Sec. 3.4 / Algorithm 1] The learning rate eta used in Eq. (2) is never reported, and it is unclear whether the same eta is used for the hypernetwork update in Algorithm 1 line 9. Please state all hyperparameter values, including eta, T, and batch sizes for each experiment.
  2. [Fig. 4 caption] The caption contains a typo: 'AHFQ' should be 'AFHQ'.
  3. [Sec. 4.2, sentence] The sentence 'we sample one of and encode' is incomplete; it should read something like 'we sample one of the 38 renderings and encode it with a pre-trained ViT.'
  4. [Sec. 4.2, Objaverse experiment] The Objaverse experiment is limited to 128 shapes and is a proof-of-concept overfitting study. This should be stated more clearly in the main text, not only as a side remark, to avoid overgeneralization.
  5. [Sec. 5, Limitations] The limitations section mentions parameter inefficiency but does not mention the potential mismatch between the simulated vanilla-GD trajectory and the actual optimizer used for the task network. Please add a discussion of this approximation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the gradient-matching objective is a self-consistency fixed-point equation with external task-loss gradients, not a fit to precomputed targets, and no load-bearing self-citations appear.

full rationale

The paper's central derivation (Eqs. 2-3) defines H_phi(x,t) as a solution of the functional recurrence H(x,t+1) = H(x,t) - eta grad L_task(H(x,t),x), with H(x,0)=theta0 enforced by construction via Eq. 4. This is a genuine fixed-point/consistency condition: the target contains the task-loss gradient evaluated at the current prediction, but that gradient is an external signal derived from the task objective and the conditioning sample, not from the hypernetwork's final output or from any precomputed optimized weight. Thus the supervision is not equivalent to the output by construction. The self-referential appearance of H_phi on both sides of Eq. 3 is a bootstrapped trajectory formulation of the kind used in consistency and flow-matching methods, not a reduction of the prediction to its input. Eq. 4's t/T offset parameterization is an ansatz whose cited inspirations ([22], [47]) are external prior works, not self-citations by the present authors, and no uniqueness theorem from the authors' own prior work is invoked to force the choice. The skeptical concern that the moving-target objective is never directly validated against a true vanilla-gradient-descent trajectory (no final Eq. 3 residual, no convergence analysis) is an empirical-validation gap and a correctness risk, not a circular reduction: nothing in the paper's equations makes the claimed output equal to its own input by definition. The ablations against a reconstruction-loss baseline also show non-trivial behavioral differences, consistent with a non-circular training signal. Therefore the circularity score is 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical entities. It relies on a small set of hand-chosen hyperparameters (eta, T, theta0) and on domain assumptions about differentiability, optimizer approximation, and capacity that are not verified analytically. The most fragile assumption is that vanilla gradient descent with a fixed learning rate approximates the true optimization of the task network.

free parameters (3)
  • eta (learning rate in Eq. 2) = not reported
    The one-step target in Eq. 2 depends on this learning rate, but the paper never states its value; it is a hand-chosen hyperparameter that directly scales the gradient-matching supervision.
  • T (maximum optimization step) = 500 (image personalization), 10000 (3D)
    The queried final state H(x,T) is defined by T; results depend on choosing a large enough T that the trajectory has converged.
  • theta0 (shared initialization for all samples) = random, fixed once
    All trajectories start from the same randomly initialized weights; this is a design choice, not a fitted parameter, but it constrains what the field can represent.
assumptions (4)
  • domain assumption The task loss L_task is differentiable with respect to all hypernetwork-produced weights, and its gradient can be computed for any theta_t = H_phi(x,t).
    Algorithm 1 evaluates ∇_theta L_task at hypernetwork outputs; this requires backpropagation through the full task network for every sample and timestep.
  • domain assumption The true task optimization is well approximated by vanilla gradient descent with a fixed learning rate eta (Eq. 2).
    DreamBooth uses Adam and a stochastic denoising objective; occupancy networks are also typically trained with adaptive optimizers. The consistency target is therefore not the true trajectory.
  • domain assumption The least-squares consistency problem has a solution that generalizes across all x and t, and the shared initialization theta0 is a suitable starting point.
    The method fixes one theta0 for all samples and relies on the network to represent all trajectories simultaneously; no expressivity or well-posedness analysis is provided.
  • domain assumption A finite T of 500 (image) or 10000 (3D) is enough for the queried final state to be a converged solution.
    No convergence check is reported; the paper uses these values as hyperparameters and applies fast fine-tuning after the image queries.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HyperNet Fields: Efficiently Training Hypernetworks without Ground Truth by Learning Weight Trajectories." pith.science (2026). https://pith.science/paper/JLDHPFMV

@misc{pith2026241217040,
  author       = {Pith},
  title        = {Pith review of: HyperNet Fields: Efficiently Training Hypernetworks without Ground Truth by Learning Weight Trajectories},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JLDHPFMV}},
  note         = {Machine review of arXiv:2412.17040}
}
read the original abstract

To efficiently adapt large models or to train generative models of neural representations, Hypernetworks have drawn interest. While hypernetworks work well, training them is cumbersome, and often requires ground truth optimized weights for each sample. However, obtaining each of these weights is a training problem of its own-one needs to train, e.g., adaptation weights or even an entire neural field for hypernetworks to regress to. In this work, we propose a method to train hypernetworks, without the need for any per-sample ground truth. Our key idea is to learn a Hypernetwork `Field` and estimate the entire trajectory of network weight training instead of simply its converged state. In other words, we introduce an additional input to the Hypernetwork, the convergence state, which then makes it act as a neural field that models the entire convergence pathway of a task network. A critical benefit in doing so is that the gradient of the estimated weights at any convergence state must then match the gradients of the original task -- this constraint alone is sufficient to train the Hypernetwork Field. We demonstrate the effectiveness of our method through the task of personalized image generation and 3D shape reconstruction from images and point clouds, demonstrating competitive results without any per-sample ground truth.

Figures

Figures reproduced from arXiv: 2412.17040 by the authors.

Figure 1
Figure 1. Teaser – We propose hypernetwork fields, where we learn the entire weight space trajectory instead of only the final state, which allows us to train without ever needing to know the final converged weights. Our method can be applied to any application of hypernetworks, including diffusion model personalization and modeling of 3D neural representations. Abstract To efficiently adapt large models or to train generativ… view at source ↗
Figure 2
Figure 2. The Hypenet Fields Framework. Left: Our Hypernetwork with gradient-based supervision from a task-specific network – dreambooth [40]. Right: The sampling process to generate personalized images with our hypernetwork framework. into the training process of the hypernet field Hϕ(t, x), as we are ‘skipping ahead’ parts of the optimization loop of Eq. (1) with the estimates coming from Hϕ(t, x). Through this amortization… view at source ↗
Figure 3
Figure 3. Qualitative examples (CelebA HQ) – Qualitative examples of personalized human face generation using the CelebA HQ dataset [21, 27] are shown. Our hypernetwork field achieves fast adaptation, producing results comparable to DreamBooth [40] and Textual Inversion [13] while preserving individual features in a personalized manner. Additional examples are provided in the supplementary material. For inference, consider Dr… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative examples (AHFQ) – Qualitative results of personalized animal image generation from the AFHQ dataset [6] demonstrate the adaptability of our hypernetwork field. Compared to DreamBooth [40] and Textual Inversion [13], our method effectively captures the speci…
Figure 6
Figure 6. Figure 6: Qualitative examples (3D Shapes) – We show example shape reconstruction results converting (left) point clouds to 3D shapes and (right) images to 3D shapes. More qualitative examples are available in the supplementary material. in the case of the AFHQ dataset (Tab. 2),…
Figure 5
Figure 5. Figure 5: Fast training (AFHQ) – We show example outcomes of our method only after 100 iterations of training, as raw output from our Hypernetwork without fast fine-tuning. As shown, even after only 100 iterations, our model is able to generate images that are visually very simi…
Figure 7
Figure 7. Figure 7: Hypernet Field output example (ShapeNet) – We show the output of our HyperNet Field for various points along the optimiza￾tion trajectory as well as the intersection over union (IoU) over all values of T. As shown, the HyperNet Field is able to accurately estimate the …
Figure 8
Figure 8. Figure 8: Celeba without fast fine tuning – Each of these im￾ages had their dreambooth parameters estimated in a single for￾ward pass of our hypernetwork field. Our model can be seen to produce reasonable results even without extra fine tuning. Condition image Instagram selfie W…
Figure 10
Figure 10. Figure 10: AFHQ without fast fine tuning – Each of these im￾ages had their dreambooth parameters estimated in a single for￾ward pass of our hypernetwork field. Our model can be seen to produce reasonable results even without extra fine tuning. Condition image In the jungle In th…
Figure 12
Figure 12. Figure 12: Example results of training with the task-specific loss – Training directly with the task-specific loss results in the hy￾pernetworks training simply learning to overfit to a specific train￾ing sample, and ignoring the user prompt. As shown, results start being irrele…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 44 canonical work pages

  1. [1]

    Learning representations and generative models for 3d point clouds

    Panos Achlioptas, Olga Diamanti, Ioannis Mitliagkas, and Leonidas Guibas. Learning representations and generative models for 3d point clouds. In ICML, 2018. 7

    work page 2018

  2. [2]

    Hyperfields: To- wards zero-shot generation of nerfs from text

    Sudarshan Babu, Richard Liu, Avery Zhou, Michael Maire, Greg Shakhnarovich, and Rana Hanocka. Hyperfields: To- wards zero-shot generation of nerfs from text. In ICML. PMLR, 2024. 2

    work page 2024

  3. [3]

    Dickson, Ryan M

    Gavin Barill, Neil G. Dickson, Ryan M. Schmidt, David I. W. Levin, and Alec Jacobson. Fast winding numbers for soups and clouds. Transactions On Graphics, 37:1–12, 2018. 8

    work page 2018

  4. [4]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Sub- biah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakan- tan, Pranav Shyam, Girish Sastry, Amanda Askell, Sand- hini Agarwal, Ariel Herbert-V oss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz...

    work page 2020

  5. [5]

    Emerg- ing properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herv ´e J´egou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers. In ICCV, pages 9650–9660, 2021. 6

    2021

  6. [6]

    Stargan v2: Diverse image synthesis for multiple domains

    Yunjey Choi, Youngjung Uh, Jaejun Yoo, and Jung-Woo Ha. Stargan v2: Diverse image synthesis for multiple domains. In CVPR, 2020. 5, 6, 11

    work page 2020

  7. [7]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. In CVPR, pages 13142– 13153, 2023. 8

    work page 2023

  8. [8]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Alexey Dosovitskiy. An image is worth 16x16 words: Trans- formers for image recognition at scale. ICLR, 2021. 5, 7

    work page 2021

Show all 53 references
  1. [9]

    Interpreting the weight space of customized dif- fusion models

    Amil Dravid, Yossi Gandelsman, Kuan-Chieh Wang, Rameen Abdal, Gordon Wetzstein, Alexei A Efros, and Kfir Aberman. Interpreting the weight space of customized dif- fusion models. arXiv preprint arXiv:2406.09413, 2024. 8

    2024 arXiv

  2. [10]

    Implicit generation and mod- eling with energy based models

    Yilun Du and Igor Mordatch. Implicit generation and mod- eling with energy based models. NeurIPS, 32, 2019. 3

    2019

  3. [11]

    Hyperdiffusion: Generating implicit neural fields with weight-space diffusion

    Ziya Erkoc ¸, Fangchang Ma, Qi Shan, Matthias Nießner, and Angela Dai. Hyperdiffusion: Generating implicit neural fields with weight-space diffusion. In ICCV, pages 14300– 14310, 2023. 2, 3, 8

    2023

  4. [12]

    One step diffusion via shortcut models

    Kevin Frans, Danijar Hafner, Sergey Levine, and Pieter Abbeel. One step diffusion via shortcut models. arXiv preprint arXiv:2410.12557, 2024. 3

    2024 arXiv

  5. [13]

    An image is worth one word: Personalizing text-to-image gen- eration using textual inversion

    Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patashnik, Amit H Bermano, Gal Chechik, and Daniel Cohen-Or. An image is worth one word: Personalizing text-to-image gen- eration using textual inversion. ICLR, 2023. 5, 6

    2023

  6. [14]

    Learning energy-based models by dif- fusion recovery likelihood

    Ruiqi Gao, Yang Song, Ben Poole, Ying Nian Wu, and Diederik P Kingma. Learning energy-based models by dif- fusion recovery likelihood. ICLR, 2021. 3

    2021

  7. [15]

    Atlasnet: A papier-m ˆach´e ap- proach to learning 3d surface generation

    Thibault Groueix, Matthew Fisher, Vladimir G Kim, Bryan Russell, and Mathieu Aubry. Atlasnet: A papier-m ˆach´e ap- proach to learning 3d surface generation. In CVPR, 2018. 7

    2018

  8. [16]

    Hypernetworks

    David Ha, Andrew Dai, and Quoc V Le. Hypernetworks. ICLR, 2017. 2

    2017

  9. [17]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. In NeurIPS, pages 6840–6851,

  10. [18]

    LoRA: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. In ICLR, 2022. 6

    2022

  11. [19]

    Mani- foldplus: A robust and scalable watertight manifold surface generation method for triangle soups

    Jingwei Huang, Yichao Zhou, and Leonidas Guibas. Mani- foldplus: A robust and scalable watertight manifold surface generation method for triangle soups. arxiv, 2020. 8

    2020

  12. [20]

    Shap-e: Generating condi- tional 3d implicit functions

    Heewoo Jun and Alex Nichol. Shap-e: Generating condi- tional 3d implicit functions. arxiv, 2023. 7

    2023

  13. [21]

    Progressive growing of gans for improved quality, stability, and variation

    Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gans for improved quality, stability, and variation. In ICLR, 2018. 5

    2018

  14. [22]

    Elucidating the design space of diffusion-based generative models

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. NeurIPS, 35:26565–26577, 2022. 4

    2022

  15. [23]

    Consistency trajectory mod- els: Learning probability flow ode trajectory of diffusion

    Dongjun Kim, Chieh-Hsin Lai, Wei-Hsiang Liao, Naoki Mu- rata, Yuhta Takida, Toshimitsu Uesaka, Yutong He, Yuki Mitsufuji, and Stefano Ermon. Consistency trajectory mod- els: Learning probability flow ode trajectory of diffusion. In ICLR. 3

  16. [24]

    A tutorial on energy-based learn- ing

    Yann LeCun, Sumit Chopra, Raia Hadsell, Marc’Aurelio Ranzato, and Fu Jie Huang. A tutorial on energy-based learn- ing. Predicting structured data, 1(2):1–31, 2006. 3

    2006

  17. [25]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maxi- milian Nickel, and Matt Le. Flow matching for generative modeling. ICLR, 2023. 3

    2023

  18. [26]

    Zero-1-to- 3: Zero-shot one image to 3d object

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tok- makov, Sergey Zakharov, and Carl V ondrick. Zero-1-to- 3: Zero-shot one image to 3d object. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9298–9309, 2023. 7

    2023

  19. [27]

    Deep learning face attributes in the wild

    Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In ICCV, 2015. 5, 11

    2015

  20. [28]

    Marching cubes: A high resolution 3d surface construction algorithm

    William E Lorensen and Harvey E Cline. Marching cubes: A high resolution 3d surface construction algorithm. InSem- inal graphics: pioneering efforts that shaped the field, pages 347–353, 1998. 8

    1998

  21. [29]

    Occupancy networks: Learning 3d reconstruction in function space

    Lars Mescheder, Michael Oechsle, Michael Niemeyer, Se- bastian Nowozin, and Andreas Geiger. Occupancy networks: Learning 3d reconstruction in function space. In CVPR, pages 4460–4470, 2019. 2, 7, 8

    2019

  22. [30]

    Deepsdf: Learning con- tinuous signed distance functions for shape representation

    Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning con- tinuous signed distance functions for shape representation. In CVPR, pages 165–174, 2019. 2

    2019

  23. [31]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In ICCV, pages 4195–4205, 2023. 5

    2023

  24. [32]

    Hypermaml: Few-shot adaptation of deep models with hypernetworks

    Marcin Przewiezlikowski, Przemyslaw Przybysz, Jacek Ta- bor, Maciej Zikeba, and Przemyslaw Spurek. Hypermaml: Few-shot adaptation of deep models with hypernetworks. Neurocomputing, 2024. 2

    2024

  25. [33]

    Pointnet: Deep learning on point sets for 3d classification and segmentation

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In CVPR, 2017. 7

    2017

  26. [34]

    Richdreamer: A generalizable normal-depth diffusion model for detail richness in text-to- 3d

    Lingteng Qiu, Guanying Chen, Xiaodong Gu, Qi Zuo, Mu- tian Xu, Yushuang Wu, Weihao Yuan, Zilong Dong, Liefeng Bo, and Xiaoguang Han. Richdreamer: A generalizable normal-depth diffusion model for detail richness in text-to- 3d. In CVPR, pages 9914–9925, 2024. 7

    2024

  27. [35]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. In ICML, 2021. 6

    2021

  28. [36]

    Physics informed deep learning (part i): Data-driven solu- tions of nonlinear partial differential equations

    Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics informed deep learning (part i): Data-driven solu- tions of nonlinear partial differential equations. arxiv, 2017. 3

    2017

  29. [37]

    Physics informed deep learning (part ii): Data-driven discov- ery of nonlinear partial differential equations

    Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics informed deep learning (part ii): Data-driven discov- ery of nonlinear partial differential equations. arxiv, 2017

    2017

  30. [38]

    Physics-informed neural networks: A deep learning frame- work for solving forward and inverse problems involving nonlinear partial differential equations

    Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics-informed neural networks: A deep learning frame- work for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computa- tional Physics, 378:686–707, 2019. 3

    2019

  31. [39]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. In CVPR, 2022. 5, 6

    2022

  32. [40]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In CVPR, pages 22500–22510, 2023. 1, 3, 4, 5, 6, 7

    2023

  33. [41]

    Hyperdreambooth: Hypernetworks for fast personalization of text-to-image models

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Wei Wei, Tingbo Hou, Yael Pritch, Neal Wadhwa, Michael Rubinstein, and Kfir Aberman. Hyperdreambooth: Hypernetworks for fast personalization of text-to-image models. InCVPR, pages 6527–6536, 2024. 2, 3, 5, 6, 7

    2024

  34. [42]

    Learning representations by back-propagating er- rors

    David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning representations by back-propagating er- rors. NATURE, 323(6088):533–536, 1986. 8

    1986

  35. [43]

    Photorealistic text-to-image diffusion models with deep language understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. NeurIPS, 35:36479–36494, 2022. 5

    2022

  36. [44]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In ICML, pages 2256– 2265, 2015. 3

    2015

  37. [45]

    Generative modeling by esti- mating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by esti- mating gradients of the data distribution. NeurIPS, 32, 2019. 3

    2019

  38. [46]

    Maximum likelihood training of score-based diffusion mod- els

    Yang Song, Conor Durkan, Iain Murray, and Stefano Ermon. Maximum likelihood training of score-based diffusion mod- els. NeurIPS, pages 1415–1428, 2021. 3

    2021

  39. [47]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. arxiv, 2023. 3, 4

    2023

  40. [48]

    A connection between score matching and denoising autoencoders

    Pascal Vincent. A connection between score matching and denoising autoencoders. Neural Computation, 23(7):1661– 1674, 2011. 3

    2011

  41. [49]

    Grewe, and Joao Sacramento

    Johannes von Oswald, Christian Henning, Benjamin F. Grewe, and Joao Sacramento. Continual learning with hy- pernetworks. In ICLR, 2019. 2

    2019

  42. [50]

    Pointflow: 3d point cloud generation with continuous normalizing flows

    Guandao Yang, Xun Huang, Zekun Hao, Ming-Yu Liu, Serge Belongie, and Bharath Hariharan. Pointflow: 3d point cloud generation with continuous normalizing flows. In ICCV, pages 4541–4550, 2019. 8

    2019

  43. [51]

    Graph hy- pernetworks for neural architecture search

    Chris Zhang, Mengye Ren, and Raquel Urtasun. Graph hy- pernetworks for neural architecture search. ICLR, 2019. 2

    2019

  44. [52]

    Reconstruction loss

    Ablations To further motivate our design choices, we show additional ablation results. Reconstruction loss. Specifically, we further compare us- ing our method against using an identical setup as ours, but directly learning the hypernetwork through the task-specific loss. More...

  45. [53]

    Specifically, we include images from the AFHQ dataset [6] sampled directly from the hypernetwork (Fig- ure 10) and after fast fine-tuning (Figure 11)

    Extra Qualitative Results We present additional results generated by our hypernet- work, both in its direct output form and after fast fine- tuning. Specifically, we include images from the AFHQ dataset [6] sampled directly from the hypernetwork (Fig- ure 10) and after fast fi...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.