Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Efficient Scaling of Diffusion Transformers for Text-to-Image Generation

T0 review · 4 major / 4 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 2.3B pure self-attention U-ViT matches SDXL's U-Net on text-image alignment at matched training steps, while running faster end-to-end.

desk verdict Large controlled DiT scaling study with real value, but the headline 'better than SDXL' overstates the body's 'matches' and the equal-step comparison is confounded by ~4x more FLOPs. read the letter →

arxiv 2412.12391 v1 pith:IWIGOUS2 submitted 2024-12-16 cs.CV cs.CLcs.LG

Hao Li , Shamit Lal , Zhiheng Li , Yusheng Xie , Ying Wang , Yang Zou , Orchid Majumder , R. Manmatha
show 4 more authors
Zhuowen Tu Stefano Ermon Stefano Soatto Ashwin Swaminathan
This is my paper · ORCID
classification cs.CVcs.CLcs.LG
keywords diffusiontransformerstext-to-imagegenerationU-ViTmodelscalingself-attentioncross-attentionlongcaptionsimageinpainting
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

Text-to-image diffusion models can be built on different backbones, and this paper claims that the simplest transformer design—U-ViT, a pure self-attention Vision Transformer where text tokens are concatenated with image tokens and refined through every layer—scales more effectively than cross-attention DiT variants. In controlled training on up to 600M image-text pairs, a 2.3B-parameter U-ViT matches SDXL's U-Net on the TIFA and ImageReward alignment metrics after 500K steps while running with lower end-to-end latency, despite roughly three times the theoretical FLOPs. The authors also report that scaling dataset size and caption information density improves alignment for all architectures, and that U-ViT gains more from added data than the U-Net baseline does. If these claims hold, the main practical consequence is that a simpler architecture with token-level text conditioning can replace a more complex cross-attention U-Net while being easier to extend to new conditions.

What carries the argument

The central object is the U-ViT backbone: noisy image latents are patchified, then concatenated with timestep and text embeddings into a single token sequence processed by stacked self-attention blocks, with long skip connections between early and late layers and no down/up sampling. Text conditioning is therefore not a separate cross-attention pass at every block but a set of tokens that move through the whole transformer and are progressively refined. This design is what the paper claims drives the scaling result, and it is also the mechanism that lets new conditions (inpainting masks, canny maps) be added by simply tokenizing and concatenating them.

What would settle it

Run the 2.3B U-ViT and the SDXL U-Net from scratch under identical data, VAE, and text encoder with at least five seeds, and compare TIFA/ImageReward at 500K steps with confidence intervals; if the U-Net's mean exceeds U-ViT by more than the seed noise, the matching claim fails. Also record end-to-end latency on the same H100 hardware at 512x512 resolution to check whether the reported latency advantage is reproducible.

Watch

Extended reading notes

Core claim

Contrary to the trend of injecting text through cross-attention blocks, the paper argues that feeding text tokens into the same self-attention stack as image tokens gives a better scaling law for text-to-image generation. Training from scratch on the LensArt and SSTK datasets with the same VAE, text encoder, and schedule, the authors scale PixArt-α, LargeDiT, and U-ViT from 0.3B to 8B parameters. They find that a 2.3B U-ViT (hidden width 2048, depth 42, 16 heads) significantly outperforms the SD2 U-Net and matches the SDXL U-Net on both TIFA and ImageReward after 500K steps; larger U-ViTs (3.1B and 3.7B) do not improve further. The explanation offered is that self-attention over condition tokens implicitly fine-tunes the text representation through the network, an effect that matters more as the backbone scales, and that long skip connections are what make in-context conditioning viable.

Load-bearing premise

The central ranking rests on single training runs at matched step counts without seeds or error bars, assuming those curves are precise enough to compare architectures and that the SD2/SDXL U-Net baselines from prior work isolate the architecture effect.

Editorial extensions

If this is right

  • A 2.3B self-attention U-ViT can be dropped into a text-to-image pipeline that otherwise keeps the SDXL VAE and text encoder, matching U-Net alignment after 500K steps with lower end-to-end latency.
  • Cross-attention DiT variants such as PixArt-α and LargeDiT are not inherently better than a pure self-attention design at comparable parameter counts; U-ViT converges faster at the 2B scale.
  • Scaling training data from 250M to 600M image-text pairs improves alignment for all backbones, and U-ViT shows a larger relative gain than SDXL U-Net, implying self-attention models benefit more from data.
  • New conditions can be added to U-ViT by token concatenation without specialized adapters, and in the paper's evaluations this beats channel concatenation and dedicated baselines for inpainting and canny conditioning.
  • Longer captions help mainly through higher information density, not raw length, so caption rewriting should target coverage of TIFA-style elements.

Reading between the lines

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

  • If the result replicates across seeds, it would suggest that theoretical FLOPs are a poor predictor of practical scaling in text-to-image backbones; end-to-end latency and per-step compute are the quantities that determine which architecture to pick.
  • The paper's explanation implies that scaling U-ViT effectively scales the text encoder too, since text tokens are refined inside the backbone; a testable extension would be to measure how much of the 2.3B model's gain comes from improved text representations rather than denoising capacity.
  • The token-concatenation conditioning scheme is demonstrated only for image inpainting and canny maps; a natural extension is to apply the same recipe to text-to-video or multi-frame conditioning, where the paper argues the design should transfer without new machinery.
  • Because the 3.1B and 3.7B U-ViTs do not beat the 2.3B model on LensArt, the scaling frontier may be compute- or data-limited rather than capacity-limited; adding more data or longer captions could make the larger models pull ahead, which the current experiments do not test.
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 / 4 minor

Summary. The paper presents a large-scale empirical study of three diffusion-transformer (DiT) families for text-to-image generation: PixArt-α, LargeDiT, and U-ViT. The authors train these architectures from scratch at scales from 0.3B to 8B parameters on proprietary datasets of up to 600M image-text pairs, using a shared autoencoder, text encoder, and training protocol. Their central finding is that U-ViT, a pure self-attention architecture with condition tokens concatenated at the input, scales more effectively than cross-attention based DiT variants and matches or exceeds a U-Net baseline in terms of TIFA and ImageReward scores at equal numbers of training steps. They also study scaling of dataset size and caption richness, propose that U-ViT's self-attention implicitly fine-tunes the text encoder, and demonstrate that token concatenation enables image inpainting and canny conditioning without architectural changes. The paper's abstract claims that a 2.3B U-ViT achieves better performance than the SDXL U-Net, while the body states that it matches SDXL U-Net after 500K steps; the data-scaling experiments show U-ViT surpassing SDXL only after adding a second large dataset.

Significance. If the empirical claims hold, the paper would provide a useful architecture-level comparison of leading DiT designs for text-to-image generation, a relatively scarce contribution given the cost of controlled large-scale training. The paper's strengths are its breadth—covering three DiT families, parameter counts up to 8B, and datasets up to 600M images—and its separation of architecture ablations from data scaling. The token-concatenation extension for inpainting and canny conditioning is a practical result that could generalize to other conditional generation tasks. The paper also makes its evaluation protocol explicit, reporting TIFA, ImageReward, and FID across many configurations. However, the headline claim of surpassing SDXL is not consistently supported by the body's 'matches SDXL' characterization, and the equal-step comparison does not control for the substantially higher per-step compute of U-ViT, which weakens the central architectural-scaling conclusion.

major comments (4)
  1. [Abstract and Sec. 3.4] The abstract states that a 2.3B U-ViT 'can get better performance than SDXL UNet,' but Sec. 3.4 says it 'matches SDXL U-Net in both TIFA and ImageReward after 500K steps,' and Fig. 13 shows a clear superiority only after adding the SSTK dataset (350M additional images). This is a load-bearing discrepancy: the controlled comparison on LensArt alone supports 'matches,' not 'better.' The abstract and conclusion should be reworded to match the evidence, or the authors should provide a specific LensArt-only configuration and training step where the 2.3B U-ViT statistically exceeds SDXL U-Net on both metrics.
  2. [Table 1 and Sec. 3.4] The 'controlled setting' comparison is an equal-step comparison, not an equal-compute comparison. Table 1 lists 0.78 TMACs for the 2.3B U-ViT versus 0.20 TMACs for the SDXL U-Net at 256x256 resolution, a factor of roughly 3.9. After 500K steps, the U-ViT has consumed about 3.9x the multiply-accumulate operations of the SDXL baseline. The observed convergence advantage is therefore consistent with the U-ViT amortizing more compute per step rather than with the self-attention design being intrinsically more scalable. To support the claim that U-ViT 'scales more effectively,' the authors need a compute-normalized comparison, for example by plotting performance against total FLOPs or by including a lower-FLOPs U-ViT variant that matches SDXL's per-step budget.
  3. [Secs. 3.2-3.4, Figs. 3-6, 13] All training curves appear to be single runs without multiple seeds or error bars. Given the small metric differences being interpreted as rankings (for example, TIFA differences around 0.01-0.02 in Fig. 6), the absence of variance information makes it impossible to assess whether the observed ordering across architectures, depths, and widths is statistically reliable. At minimum, the authors should provide repeated-seed estimates for the headline comparisons: 2.3B U-ViT versus SDXL U-Net in Fig. 5, the 0.6B and 2.3B comparisons in Fig. 6, and the data-scaling comparison in Fig. 13.
  4. [Sec. 6.3 and Fig. 14] The 'information density' explanation is measured by phrase matching of TIFA element types, which is the same taxonomy used to construct TIFA questions and to compute the TIFA score. The observation that captions with more matched TIFA phrases lead to better TIFA scores is therefore partly self-referential: it shows that the training captions contain more of what TIFA tests, not independently that information density per se improves alignment. The conclusion in Sec. 6.3 needs a measure of caption informativeness that is not derived from the evaluation metric, such as a human study, a KL-divergence based caption-category statistic, or a downstream retrieval-based measure, to break the circularity.
minor comments (4)
  1. [Sec. 1 and Sec. 3] There are several typos: 'rigors ablations' should be 'rigorous ablations,' 'archtiecture' should be 'architecture,' and 'unertand' in Sec. 6.3 should be 'understand.'
  2. [Sec. 4.1] The phrase 'the condition tokens are contacted with timestep embedding' should read 'concatenated with timestep embedding.'
  3. [Sec. 5.2 and Appendix B] The definition and evaluation of the modified metric TIFA-COCO are important for the inpainting and canny results, but the paper does not state the number of questions or the VQA model used for this variant in the main text; the appendix should be referenced more explicitly when the metric is introduced.
  4. [Sec. 3.1] The evaluation section says DDIM is used with a fixed seed, but does not state whether the seed is varied across the 4K/10K prompts; please clarify whether the 'fixed seed' applies to the sampler stochasticity and whether results are averaged over any seed choices.

Circularity Check

1 steps flagged · score 2.0 of 10

Central scaling claims are empirical and self-contained; only the Sec 6.3 'information density' explanation is partially self-referential because it uses TIFA's own element phrases to explain TIFA gains.

  1. self definitional [Section 6.3, Figure 14 and surrounding text]
    "Based on element phrases (note that each element can contain more than one English word) in TIFA, we do phrase matching for (1) LensArt original caption, (2) LensArt long caption, and (3) SSTK captions. ... The results show that both LensArt long captions and SSTK have higher percentage of captions with matched TIFA element phrases compared to LensArt caption in almost all TIFA element types, which explains the relative performance difference in Fig. 12."

    The paper's 'information density' explanation is operationalized by matching the same TIFA element phrases that TIFA uses to generate its VQA questions. TIFA scores a caption-to-image pair by asking questions about elements extracted from that caption; a caption containing more TIFA element phrases therefore has more elements available to be queried. Declaring that captions with more matched TIFA element phrases 'explain' higher TIFA scores is partly a restatement of the metric's construction rather than an independent cause. The explanation retains some independent content because the generated image must actually depict the elements for VQA to answer correctly, so the circularity is partial and confined to this explanatory section.

full rationale

The paper is primarily an empirical scaling study. The headline architecture claims in Secs 3.2-3.4 and Figs 3-6, 13 are comparisons against externally defined metrics (TIFA, ImageReward, FID) and against baselines trained in prior work; no parameter is fitted to the target metric and then renamed as a prediction. The use of the LensArt/SSTK datasets and the SD2/SDXL U-Net baselines from Li et al. 2024a involves self-citation, but it is not load-bearing in a circular sense: those are external empirical resources, and the paper's conclusions do not reduce to assertions in the cited prior paper. The only notable circular element is the Sec 6.3 explanatory claim, where 'information density' is measured by phrase matching against TIFA's own element taxonomy and then used to explain TIFA improvements; that is a partial self-definitional explanation. Separately, the equal-step comparison in Sec 3.4 and Table 1 compares a 2.3B U-ViT (0.78 TMACs at 256) with SDXL U-Net (0.20 TMACs), entangling architecture effects with roughly 3.9x per-step compute; this is a correctness or fairness concern rather than a circularity, so it does not raise the circularity score. Overall circularity is therefore low, with one partial self-referential explanation in a peripheral section, giving a score of 2.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claims rest on the evaluation metrics, the single-run experimental protocol, the equal-steps comparison, the reuse of prior baselines, and the proprietary datasets; each is an assumption the reader must accept without independent verification.

assumptions (5)
  • domain assumption TIFA, ImageReward, and FID are valid proxies for text-image alignment and generation quality.
    All central comparative claims are evaluated with these metrics (Sec 3.1, Appendix C); if they are biased toward certain architectures, the ranking conclusions could change.
  • ad hoc to paper A single training run per configuration, without seeds or error bars, yields reliable performance rankings.
    Every scaling figure plots one curve per model; no variance or significance information is given, yet pairwise differences like 0.80 vs 0.78 TIFA are treated as meaningful (Fig 3).
  • ad hoc to paper Equal training steps with unequal compute is the correct controlled comparison for architecture scaling.
    The paper fixes steps at 600K while GMACs differ by up to 3x across models (Table 1), so architecture and compute are conflated; this choice affects the central claim that U-ViT 'scales more effectively'.
  • domain assumption Baselines from prior work (Li et al. 2024a) were trained under settings identical to this paper's runs.
    SD2-UNet and SDXL-UNet curves are reused from the same-group CVPR paper; the fairness of the comparison depends on identical data, optimizer, and step counts.
  • domain assumption LensArt and SSTK are representative high-quality text-image datasets, despite not being publicly released.
    Appendix A describes filtering but the data are proprietary; scaling conclusions may not transfer to other data distributions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Scaling of Diffusion Transformers for Text-to-Image Generation." pith.science (2026). https://pith.science/paper/IWIGOUS2

@misc{pith2026241212391,
  author       = {Pith},
  title        = {Pith review of: Efficient Scaling of Diffusion Transformers for Text-to-Image Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IWIGOUS2}},
  note         = {Machine review of arXiv:2412.12391}
}
read the original abstract

We empirically study the scaling properties of various Diffusion Transformers (DiTs) for text-to-image generation by performing extensive and rigorous ablations, including training scaled DiTs ranging from 0.3B upto 8B parameters on datasets up to 600M images. We find that U-ViT, a pure self-attention based DiT model provides a simpler design and scales more effectively in comparison with cross-attention based DiT variants, which allows straightforward expansion for extra conditions and other modalities. We identify a 2.3B U-ViT model can get better performance than SDXL UNet and other DiT variants in controlled setting. On the data scaling side, we investigate how increasing dataset size and enhanced long caption improve the text-image alignment performance and the learning efficiency.

Figures

Figures reproduced from arXiv: 2412.12391 by the authors.

Figure 1
Figure 1. Examples of high-resolution images generated by a 2.3B U-ViT 1K model. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the design of SDXL U-Net, DiT (e.g., PixArt- [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Scaling PixArt-α on the depth and width dimensions. 3.3 SCALING LARGEDIT We further employ LargeDiT (Gao et al., 2024) as the denoising backbone and explore its scaled version. The original LargeDiT comes with 0.6B, 3B1 , and 7B pre-trained versions. We ablate LargeDiT in the dimension of depth, hidden dimension, and number of heads. As shown in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Scaling LargeDiT variants from 1.7B to 7.6B. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Scaling U-ViT along hidden dimension h, depth d and combined dimensions on LensArt. compare PixArt-α, LargeDiT and U-ViT in similar architecture settings. Specifically, we compare them in the configurations of original DiT-XL (0.6B) and their scaled versions at the 2B …
Figure 6
Figure 6. Figure 6: Comparing different DiT designs in similar architecture hyperparameters at different scales. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: (a) The effect of removing downsampling and using fixed channels in U-Net. (b) The effect [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: (a) Comparing fine-tuning and freezing text encoder during training for cross-attention [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Channel and token concatenation for extending U-ViT to image inpainting. (a) Concatenat [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: a) Comparison of token and channel concatenation approaches for image inpainting on [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: The distribution of caption length (number of tokens) of LensArt original caption, LensArt long caption, and SSTK caption. 50000 75000 100000 125000 150000 175000 200000 225000 250000 Training Steps 0.600 0.625 0.650 0.675 0.700 0.725 0.750 0.775 TIFA Score LensArt, 2…
Figure 13
Figure 13. Figure 13: The scalability of U-ViT and SD models trained with larger datasets. Solid lines are on [PITH_FULL_IMAGE:figures/full_fig_p010_13.png]
Figure 14
Figure 14. Figure 14: Percentage of captions with matched TIFA element phrases in each TIFA element type. [PITH_FULL_IMAGE:figures/full_fig_p010_14.png]
Figure 15
Figure 15. Figure 15: Scaling PixArt-α on the depth and width dimensions in terms of TIFA, ImageReward and FID. The first row is scaling along 100000 200000 300000 400000 500000 600000 Training Steps 0.65 0.70 0.75 0.80 0.85 TIFA Score LensArt, 256, SDXL-AE, CFG 7.5, BS=2048 SD2-UNet (0.87…
Figure 16
Figure 16. Figure 16: Scaling LargeDiT variants from 1.7B to 8B and their performance in TIFA, ImageReward [PITH_FULL_IMAGE:figures/full_fig_p016_16.png]
Figure 17
Figure 17. Figure 17: Scaling U-ViT along width h, depth d and combined dimensions in terms of TIFA, ImageReward and FID. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_17.png]
Figure 18
Figure 18. Figure 18: Comparing different DiT designs in similar architecture hyperparameters at different [PITH_FULL_IMAGE:figures/full_fig_p017_18.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Importance-Aware OBS Pruning for Diffusion Models

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Injecting spatial importance maps (e.g., CFG delta) into the OBS Hessian improves subject preservation in pruned diffusion models at high sparsity, but gains over the baseline are small and without error bars.

  2. Phase-Aligned RoPE for Mixed-Resolution Diffusion Transformer

    cs.CV 2025-11 conditional novelty 6.0 of 10

    Expressing all RoPE positions on the query's grid ('one attention, one scale') plus a small boundary content-exchange step restores mixed-resolution diffusion generation that naive position interpolation destroys.

Reference graph

Works this paper leans on

22 extracted references · 6 canonical work pages · cited by 2 Pith papers

  1. [2]

    [Online; accessed 4-March-2024]

    URL https://en.wikipedia.org/w/index.php?title=Peak_signal-to-noise_ratio& oldid=1210897995. [Online; accessed 4-March-2024]. 14 Katherine Crowson, Stefan Andreas Baumann, Alex Birch, Tanishq Mathew Abraham, Daniel Z Kaplan, and Enrico Shippole. Scalable high-resolution pixel-space image synthesis with hourglass diffusion transformers. In ICML,

    work page 2024

  2. [3]

    Scaling rectified flow transformers for high-resolution image synthesis

    1, 2 Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In Forty-first International Conference on Machine Learning, 2024a. 9, 14 Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahi...

    arXiv

  3. [6]

    URL https://doi.org/10.5281/zenodo

    doi: 10.5281/zenodo.5143773. URL https://doi.org/10.5281/zenodo. 5143773. 3 Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. Mistral 7b. arXiv preprint arXiv:2310.06825,

    arXiv

  4. [7]

    Scaling laws for neural language models

    8, 14 Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models. arXiv preprint arXiv:2001.08361,

    arXiv 2001

  5. [8]

    Manmatha, Ashwin Swaminathan, Zhuowen Tu, Stefano Ermon, and Stefano Soatto

    9 Hao Li, Yang Zou, Ying Wang, Orchid Majumder, Yusheng Xie, R. Manmatha, Ashwin Swaminathan, Zhuowen Tu, Stefano Ermon, and Stefano Soatto. On the scalability of diffusion-based text-to-image generation. In CVPR, 2024a. 3, 5, 6, 14 Zhimin Li, Jianwei Zhang, Qin Lin, Jiangfeng Xiong, Yanxin Long, Xinchi Deng, Yingfang Zhang, Xingchao Liu, Minbin Huang, Ze...

    arXiv

  6. [9]

    Sdxl: improving latent diffusion models for high-resolution image synthesis

    1, 2, 6 Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952,

    arXiv

  7. [10]

    Hierarchical text-conditional image generation with clip latents

    2, 5, 6 Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 1(2):3,

    arXiv

  8. [11]

    U-net: Convolutional networks for biomedical image segmentation

    2, 3, 5 Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th interna- tional conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, pp. 234–241. Springer,

    work page 2015

Show all 22 references
  1. [14]

    wikipedia.org/w/index.php?title=Mean_squared_error&oldid=1207422018

    URL https://en. wikipedia.org/w/index.php?title=Mean_squared_error&oldid=1207422018. [On- line; accessed 4-March-2024]. 14 Chenfei Wu, Lun Huang, Qianxi Zhang, Binyang Li, Lei Ji, Fan Yang, Guillermo Sapiro, and Nan Duan. Godiva: Generating open-domain videos from natural desc...

    2024 arXiv

  2. [15]

    Human preference score v2: A solid benchmark for evaluating human preferences of text-to-image synthesis

    14 Xiaoshi Wu, Yiming Hao, Keqiang Sun, Yixiong Chen, Feng Zhu, Rui Zhao, and Hongsheng Li. Human preference score v2: A solid benchmark for evaluating human preferences of text-to-image synthesis. arXiv preprint arXiv:2306.09341,

    arXiv

  3. [16]

    Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models

    3, 14 Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721,

    arXiv

  4. [17]

    Pytorch fsdp: experiences on scaling fully sharded data parallel

    7 Yanli Zhao, Andrew Gu, Rohan Varma, Liang Luo, Chien-Chin Huang, Min Xu, Less Wright, Hamid Shojanazeri, Myle Ott, Sam Shleifer, et al. Pytorch fsdp: experiences on scaling fully sharded data parallel. arXiv preprint arXiv:2304.11277,

    arXiv

  5. [18]

    Lumina-next: Making lumina-t2x stronger and faster with next-dit

    3 Le Zhuo, Ruoyi Du, Han Xiao, Yangguang Li, Dongyang Liu, Rongjie Huang, Wenze Liu, Lirui Zhao, Fu-Yun Wang, Zhanyu Ma, et al. Lumina-next: Making lumina-t2x stronger and faster with next-dit. arXiv preprint arXiv:2406.18583,

    arXiv

  6. [19]

    LensArt consists of 250 million image-text pairs, carefully selected from an initial pool of 1 billion noisy web image-text pairs

    2 13 A D ATASET DETAILS We use two proprietary datasets named LensArt and SSTK (Li et al., 2024a). LensArt consists of 250 million image-text pairs, carefully selected from an initial pool of 1 billion noisy web image-text pairs. SSTK contains approximately 350 million cleaned...

    2023

  7. [20]

    B.1 B ENCHMARKS We outline the evaluation benchmarks used to assess the performance of our image inpainting and canny edge conditioning models

    and SD3-Medium- Canny-ControlNet (Esser et al., 2024a) respectively. B.1 B ENCHMARKS We outline the evaluation benchmarks used to assess the performance of our image inpainting and canny edge conditioning models. TIFA TIFA benchmark has a set of prompts, and a set of question-...

    2000

  8. [21]

    as they align with human perception. • Masked Region Preservation: Peak Signal-to-Noise Ratio (PSNR) (contributors, 2024), Learned Perceptual Image Patch Similarity (LPIPS) (Zhang et al., 2018), and Mean Squared Error (MSE) (Wikipedia contributors,

    2024

  9. [2014]

    In addition to TIFA and ImageReward, we also provide the FID score, which measures the fidelity or similarity of the generated images to the groundtruth images

    2014 validation set, we name it MSCOCO-10K. In addition to TIFA and ImageReward, we also provide the FID score, which measures the fidelity or similarity of the generated images to the groundtruth images. The score is calculated based on the MSCOCO-10K prompts and their corres...

    2014

  10. [2020]

    Llama: Open and efficient foundation language models

    3 12 Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971,

    arXiv

  11. [2021]

    Tifa: Accurate and interpretable text-to-image faithfulness evaluation with question answering

    9 Yushi Hu, Benlin Liu, Jungo Kasai, Yizhong Wang, Mari Ostendorf, Ranjay Krishna, and Noah A Smith. Tifa: Accurate and interpretable text-to-image faithfulness evaluation with question answering. arXiv preprint arXiv:2303.11897,

    arXiv

  12. [2022]

    Denoising diffusion implicit models

    14 Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502,

    2010 arXiv

  13. [2023]

    Lora: Low-rank adaptation of large language models

    7 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. arXiv preprint arXiv:2106.09685,

    arXiv

  14. [2024]

    1 Junsong Chen, Chongjian Ge, Enze Xie, Yue Wu, Lewei Yao, Xiaozhe Ren, Zhongdao Wang, Ping Luo, Huchuan Lu, and Zhenguo Li

    URL https://openai.com/research/ video-generation-models-as-world-simulators . 1 Junsong Chen, Chongjian Ge, Enze Xie, Yue Wu, Lewei Yao, Xiaozhe Ren, Zhongdao Wang, Ping Luo, Huchuan Lu, and Zhenguo Li. Pixart-σ: Weak-to-strong training of diffusion transformer for 4k text-to...

    arXiv

Pith tools

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