Pith. sign in

REVIEW 4 major objections 4 minor 52 references

Stable Diffusion Models are Secretly Good at Visual In-Context Learning

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper shows that a frozen, off-the-shelf Stable Diffusion model can be repurposed for visual in-context learning through an in-place recomputation of self-attention, with no fine-tuning.

desk verdict A plausible and useful training-free V-ICL method built on attention recomputation in frozen Stable Diffusion, but the key equations and a causality control are missing from the current text. read the letter →

arxiv 2508.09949 v1 pith:AIRNWGTE submitted 2025-08-13 cs.CV cs.LG

Trevine Oorloff , Vishwanath Sindagi , Wele Gedara Chaminda Bandara , Ali Shafahi , Amin Ghiasi , Charan Prakash , Reza Ardekani This is my paper
classification cs.CVcs.LG
keywords visualin-contextlearningStableDiffusionattentionre-computationtraining-freeadaptationpromptensemblingforegroundsegmentationmodelszero-shottasktransfer
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 sets out to show that visual in-context learning does not require specialized training or task-specific datasets. Its claim is that a pretrained Stable Diffusion model, used without any fine-tuning, can infer a task from a few example source–target image pairs and apply it to a new query image, as long as the self-attention layers recompute their attention with the prompt features included. The authors call this pipeline SD-VICL and demonstrate it on foreground segmentation, single object detection, semantic segmentation, keypoint detection, edge detection, and colorization, reporting a foreground-segmentation mIoU gain of 8.9% over Visual Prompting and 3.2% over IMProv on Pascal-5i. The reason this would matter is that it turns a single frozen generative model into a general-purpose visual task learner whose task adaptation comes from the prompt and an inference-time attention change, not from weight updates.

What carries the argument

The mechanism is an in-place attention re-computation inside Stable Diffusion's self-attention layers. Instead of training the model to understand prompt grids, the method takes the frozen weights and, during the forward pass, re-runs the attention operation after the prompt's source and target features are added to the key/value context, so the query's representation is rewritten by the example. That single change carries the task structure from the prompt to the query. Prompt ensembling is handled in the latent space with implicitly weighted attention rather than averaging decoded images, avoiding the resolution loss of image-space compositing.

What would settle it

On Pascal-5i foreground segmentation, run SD-VICL twice: once with correct source-target prompt pairs and once with the target masks randomly swapped between prompts. If the mIoU stays roughly unchanged, the model is not reading task structure from the prompt and the claimed attention-based transfer is not happening.

Watch

Extended reading notes

Core claim

The central discovery is that the self-attention layers of a frozen Stable Diffusion model already contain the information needed for dense visual task transfer. The paper's SD-VICL pipeline reformulates attention during inference so that the query features are recomputed with the example prompt inserted as context, giving the query direct access to the prompt's source-target relationship. This in-place recomputation, without weight updates or auxiliary data, lets the same model switch between six tasks. The authors additionally introduce implicitly weighted prompt ensembling in latent space, which lets multiple example prompts contribute to the prediction and often improves accuracy.

Load-bearing premise

The load-bearing premise is that Stable Diffusion's self-attention features contain enough dense semantic correspondence between the query and the prompt that re-running attention with the prompt injected transfers the task to the query; without that correspondence, the method fails regardless of benchmark scores.

Editorial extensions

If this is right

  • A frozen Stable Diffusion model can serve as a general task learner: a new task requires only a few image pairs, not a fine-tuning run.
  • The same attention recomputation generalizes across six dense prediction tasks, so task-specific output heads are not needed.
  • Multiple example prompts can be combined at the feature level, with the combination weights set implicitly by attention rather than by fixed averaging.
  • Prompt consistency matters: when example labels disagree, as the authors observe on COCOStuff, ensembling can lower performance, making example selection part of practical use.

Reading between the lines

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

  • My inference: if attention recomputation is doing the work, this recipe should be portable to other frozen diffusion backbones, such as video or image-editing models, for the same training-free adaptation.
  • My inference: ablating individual self-attention layers would reveal where task transfer happens, separating early spatial-correspondence layers from later refinement layers.
  • My inference: the observed sensitivity to prompt quality implies that prompt selection is a controllable performance lever, analogous to example ordering in language-model in-context learning.
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 proposes SD-VICL, a training-free visual in-context learning method built on an off-the-shelf Stable Diffusion model. The central idea is an in-place recomputation of self-attention that injects context from example prompt images into the query image, allowing the frozen diffusion model to perform six dense prediction tasks: foreground segmentation, single object detection, semantic segmentation, keypoint detection, edge detection, and colorization. The authors report improvements over Visual Prompting and IMProv, e.g., +8.9% and +3.2% mIoU on Pascal-5i foreground segmentation, and introduce a latent-space prompt ensembling mechanism. The available manuscript text, however, does not actually present the core attention reformulation, and the experimental section is largely missing from the supplied text, making it impossible to verify reproducibility or the mechanism underlying the claims.

Significance. If the central claim holds, this is a significant result: it would demonstrate emergent visual in-context learning in a frozen text-to-image diffusion model, without any fine-tuning or task-specific data. The breadth of tasks and the comparison against strong inpainting-based baselines are valuable, and the prompt-ensembling idea is a useful addition. The claim is plausible given prior work on cross-image attention in diffusion models, but the manuscript as available does not provide the evidence needed to establish it. The value of the paper depends on closing the reproducibility and causal-inference gaps described below.

major comments (4)
  1. [Section 2] The central contribution—the in-place attention re-computation—is asserted but never specified in the submitted text. After Sec. 2.1 motivates the approach, the manuscript jumps to references without providing the equations or pseudocode for how query-image and example-prompt tokens are combined in self-attention, which layers are modified, or at which denoising steps recomputation is applied. Since every reported result depends on this formulation, the paper is not reproducible and the mechanism cannot be checked. Please provide the full attention formulation (Q/K/V construction, masking, normalization, and the exact layer/step schedule), or a precise pointer to the appendix where it appears.
  2. [Section 3 / Tables 1-4] No prompt-causality control is reported. The central claim is that the model infers the task from the example prompt, but the aggregate benchmark numbers are also consistent with the model copying the prompt's target map onto query locations via latent nearest-neighbor matching, or with SD's strong image prior doing much of the work. Please add ablations that perturb the prompt target (random masks, shifted keypoint heatmaps, wrong edge maps) while keeping the query and source image fixed, and report the effect on predictions and metrics. Also include a control where the prompt source image is removed or mismatched. Without such evidence, the gains over Visual Prompting/IMProv do not establish task inference.
  3. [Appendix C, Edge detection] The evaluation uses HED soft edge maps as pseudo-ground truth because the NYUDv2 validation set has no ground-truth edges. This makes the edge-detection results a measure of agreement with a particular trained edge detector, not edge-detection quality, and using LPIPS on soft edge maps is not standard. Either re-evaluate on a dataset with human edge annotations (e.g., BIPED or BSDS) or clearly relabel the results as HED-alignment rather than edge-detection performance.
  4. [Section 3, experimental setup] The method is training-free, but it is not parameter-free. Denoising steps, classifier-free guidance scale, the subset of attention layers and the recomputation schedule, and the Gaussian variances for keypoint heatmaps are all hand-set. None of these values are reported in the available text, and no sensitivity analysis is provided. Without this information, the reader cannot tell whether the reported gains are robust or the result of favorable settings. Please report all hyperparameters and include a sensitivity study at least for CFG scale, number of denoising steps, and attention-layer subset.
minor comments (4)
  1. [Section 1, contributions] The claim of being the 'first training-free method' for visual in-context learning is strong. Prior zero-shot diffusion-based segmentation and keypoint methods (e.g., [34], [15]) should be discussed more carefully to position the novelty.
  2. [Section 2, prompt ensembling] The 'implicitly-weighted' prompt ensembling is contrasted with SegGPT's uniform feature averaging, but the weighting mechanism is not formalized. A precise formula is needed to understand how multiple prompts are combined in the attention layers.
  3. [General] The submitted text refers to Tables 5-7 and numerous qualitative figures, but those tables and the main experiments section are not included in the available manuscript. Please ensure the main paper or supplement contains all tables, dataset split details, and evaluation protocols.
  4. [Figure 2] The qualitative example showing the same prompt as the query is suggestive, but it is not a quantitative control. Consider adding a self-prompt row to the main tables to quantify how much the model leverages the prompt when it exactly matches the query.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity demonstrated: the method is training-free, evaluated against external baselines, and makes no fitted-parameter or self-citation chain that forces the reported results.

full rationale

The paper's central claim is that an off-the-shelf Stable Diffusion model can perform visual in-context learning via an in-place attention re-computation, without fine-tuning. The reported improvements (e.g., +8.9% mIoU over Visual Prompting on Pascal-5i) are benchmark comparisons against external methods, not reconstructions of fitted values. The attention re-computation is the proposed mechanism rather than a quantity defined in terms of the target metric, so there is no self-definitional equivalence. The reference list contains no papers by the present authors, so no load-bearing self-citation or imported uniqueness theorem is present. The method's reliance on pre-existing attention-editing ideas (cross-image attention, MasaCtrl) is a normal citation of external prior work, not an ansatz smuggled in via self-citation. The available text omits the equations of Sec. 2, so the exact reformulation cannot be checked from this extract; this is an evidence gap, but the rules require exhibiting a specific reduction to claim circularity, and none is shown. The absence of a prompt-causality control is a validity concern about whether the model truly 'infers the task,' but it is not a circularity of the derivation. Overall, no circular step is identifiable from the provided text.

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

The paper introduces no new physical or architectural entity. Its assumptions are about the representational content of pretrained SD features and the validity of grid and latent ensembling protocols. The main free parameters are inference-time and prompt-construction choices, none of which are reported as fixed in the visible text.

free parameters (4)
  • Denoising steps at inference = not reported
    Diffusion inference requires choosing the number of steps; results likely depend on this choice.
  • Classifier-free guidance scale = not reported
    SD outputs are sensitive to guidance scale; no value is given in the abstract or visible methods.
  • Attention layer subset and recomputation schedule = not reported
    Which self-attention layers are modified and with what intensity is a design choice that affects task transfer.
  • Keypoint heatmap Gaussian variances = smaller for facial, larger for body keypoints
    Hand-chosen variances for the keypoint prompt construction in Sec. C; this directly affects keypoint evaluation.
assumptions (3)
  • domain assumption Stable Diffusion's self-attention features encode dense semantic correspondence between images
    The method's effectiveness rests on attention recomputation transferring task structure from prompt to query; this premise is unproven and enters in Sec. 2.
  • domain assumption Grid-style placement of source-target prompts and masked query is a valid task specification for SD
    The pipeline inherits the Visual Prompting grid canvas convention without deriving why SD can parse it.
  • domain assumption Latent-space ensembling of multiple prompts preserves and improves task inference
    The benefit of implicit weighting in attention layers is asserted, not derived from first principles.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Stable Diffusion Models are Secretly Good at Visual In-Context Learning." pith.science (2026). https://pith.science/paper/AIRNWGTE

@misc{pith2026250809949,
  author       = {Pith},
  title        = {Pith review of: Stable Diffusion Models are Secretly Good at Visual In-Context Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AIRNWGTE}},
  note         = {Machine review of arXiv:2508.09949}
}
read the original abstract

Large language models (LLM) in natural language processing (NLP) have demonstrated great potential for in-context learning (ICL) -- the ability to leverage a few sets of example prompts to adapt to various tasks without having to explicitly update the model weights. ICL has recently been explored for computer vision tasks with promising early outcomes. These approaches involve specialized training and/or additional data that complicate the process and limit its generalizability. In this work, we show that off-the-shelf Stable Diffusion models can be repurposed for visual in-context learning (V-ICL). Specifically, we formulate an in-place attention re-computation within the self-attention layers of the Stable Diffusion architecture that explicitly incorporates context between the query and example prompts. Without any additional fine-tuning, we show that this repurposed Stable Diffusion model is able to adapt to six different tasks: foreground segmentation, single object detection, semantic segmentation, keypoint detection, edge detection, and colorization. For example, the proposed approach improves the mean intersection over union (mIoU) for the foreground segmentation task on Pascal-5i dataset by 8.9% and 3.2% over recent methods such as Visual Prompting and IMProv, respectively. Additionally, we show that the proposed method is able to effectively leverage multiple prompts through ensembling to infer the task better and further improve the performance.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 40 canonical work pages

  1. [1]

    Cross-image attention for zero- shot appearance transfer

    Yuval Alaluf, Daniel Garibi, Or Patashnik, Hadar Averbuch- Elor, and Daniel Cohen-Or. Cross-image attention for zero- shot appearance transfer. In ACM SIGGRAPH 2024 Confer- ence Papers, pages 1–12, 2024. 3, 4, 15

    work page 2024

  2. [2]

    Sequential modeling enables scalable learn- ing for large vision models

    Yutong Bai, Xinyang Geng, Karttikeya Mangalam, Amir Bar, Alan L Yuille, Trevor Darrell, Jitendra Malik, and Alexei A Efros. Sequential modeling enables scalable learn- ing for large vision models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 22861–22872, 2024. 6, 7, 11, 12, 13, 15

    work page 2024

  3. [3]

    Visual prompting via image inpaint- ing

    Amir Bar, Yossi Gandelsman, Trevor Darrell, Amir Glober- son, and Alexei Efros. Visual prompting via image inpaint- ing. Advances in Neural Information Processing Systems , 35:25005–25017, 2022. 1, 2, 3, 5, 6, 7, 11, 12, 13, 18, 19, 20, 21, 22, 23

    work page 2022

  4. [4]

    In- structpix2pix: Learning to follow image editing instructions

    Tim Brooks, Aleksander Holynski, and Alexei A Efros. In- structpix2pix: Learning to follow image editing instructions. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 18392–18402, 2023. 12

    work page 2023

  5. [5]

    Lan- guage models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Sub- biah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakan- tan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Lan- guage models are few-shot learners. Advances in neural in- formation processing systems, 33:1877–1901, 2020. 1, 11

    work page 1901

  6. [6]

    Masactrl: Tuning-free mu- tual self-attention control for consistent image synthesis and editing

    Mingdeng Cao, Xintao Wang, Zhongang Qi, Ying Shan, Xi- aohu Qie, and Yinqiang Zheng. Masactrl: Tuning-free mu- tual self-attention control for consistent image synthesis and editing. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 22560–22570, 2023. 3

    work page 2023

  7. [7]

    Palm: Scaling language modeling with pathways

    Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. Palm: Scaling language modeling with pathways. Journal of Machine Learning Research, 24(240): 1–113, 2023. 11

    work page 2023

  8. [8]

    The cityscapes dataset for semantic urban scene understanding

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. In Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. 5, 12

    work page 2016

Show all 52 references
  1. [9]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  2. [10]

    Taming transformers for high-resolution image synthesis

    Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12873–12883, 2021. 11

    2021

  3. [11]

    Explore in-context learning for 3d point cloud understanding

    Zhongbin Fang, Xiangtai Li, Xia Li, Joachim M Buhmann, Chen Change Loy, and Mengyuan Liu. Explore in-context learning for 3d point cloud understanding. Advances in Neu- ral Information Processing Systems, 36, 2024. 11

    2024

  4. [12]

    Openllama: An open reproduc- tion of llama, 2023

    Xinyang Geng and Hao Liu. Openllama: An open reproduc- tion of llama, 2023. 12

    2023

  5. [13]

    Lan- guage models are general-purpose interfaces

    Yaru Hao, Haoyu Song, Li Dong, Shaohan Huang, Zewen Chi, Wenhui Wang, Shuming Ma, and Furu Wei. Lan- guage models are general-purpose interfaces. arXiv preprint arXiv:2206.06336, 2022. 1, 11

    2022 arXiv

  6. [14]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross Girshick. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 16000– 16009, 2022. 11

    2022

  7. [15]

    Unsupervised keypoints from pretrained diffusion models

    Eric Hedlin, Gopal Sharma, Shweta Mahajan, Xingzhe He, Hossam Isack, Abhishek Kar, Helge Rhodin, Andrea Tagliasacchi, and Kwang Moo Yi. Unsupervised keypoints from pretrained diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition...

    2024

  8. [16]

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

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. Advances in neural information processing systems , 30, 2017. 6, 12

    2017

  9. [17]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022. 4

    2022 arXiv

  10. [18]

    Arbitrary style transfer in real-time with adaptive instance normalization

    Xun Huang and Serge Belongie. Arbitrary style transfer in real-time with adaptive instance normalization. In Proceed- ings of the IEEE international conference on computer vi- sion, pages 1501–1510, 2017. 4

    2017

  11. [19]

    An edit friendly ddpm noise space: Inversion and manipulations

    Inbar Huberman-Spiegelglas, Vladimir Kulikov, and Tomer Michaeli. An edit friendly ddpm noise space: Inversion and manipulations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12469– 12478, 2024. 3, 4

    2024

  12. [20]

    Fss-1000: A 1000-class dataset for few- shot segmentation

    Xiang Li, Tianhan Wei, Yau Pun Chen, Yu-Wing Tai, and Chi-Keung Tang. Fss-1000: A 1000-class dataset for few- shot segmentation. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 2869–2878, 2020. 2

    2020

  13. [21]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceeding...

    2014

  14. [22]

    Explicit visual prompting for low-level structure segmenta- tions

    Weihuang Liu, Xi Shen, Chi-Man Pun, and Xiaodong Cun. Explicit visual prompting for low-level structure segmenta- tions. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 19434–19445,

  15. [23]

    Instaflow: One step is enough for high-quality diffusion- based text-to-image generation

    Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, et al. Instaflow: One step is enough for high-quality diffusion- based text-to-image generation. In The Twelfth International Conference on Learning Representations, 2023. 17

    2023

  16. [24]

    Deepfashion: Powering robust clothes recognition and retrieval with rich annotations

    Ziwei Liu, Ping Luo, Shi Qiu, Xiaogang Wang, and Xiaoou Tang. Deepfashion: Powering robust clothes recognition and retrieval with rich annotations. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 1096–1104, 2016. 6, 12 9

    2016

  17. [25]

    Localizing object-level shape variations with text-to-image diffusion models

    Or Patashnik, Daniel Garibi, Idan Azuri, Hadar Averbuch- Elor, and Daniel Cohen-Or. Localizing object-level shape variations with text-to-image diffusion models. In Proceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 23051–23061, 2023. 3

    2023

  18. [26]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  19. [27]

    Scaling language models: Methods, analysis & insights from train- ing gopher

    Jack W Rae, Sebastian Borgeaud, Trevor Cai, Katie Milli- can, Jordan Hoffmann, Francis Song, John Aslanides, Sarah Henderson, Roman Ring, Susannah Young, et al. Scaling language models: Methods, analysis & insights from train- ing gopher. arXiv preprint arXiv:2112.11446, 2021. 11

    2021 arXiv

  20. [28]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2, 3, 11

    2022

  21. [29]

    Imagenet large scale visual recognition challenge

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, San- jeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge. International journal of computer vision, 115:211–252, 2015. 6, 12

    2015

  22. [30]

    Laion-5b: An open large-scale dataset for training next generation image-text models

    Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Worts- man, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. Advances in Neural In- f...

    2022

  23. [31]

    One-shot learning for semantic segmentation

    Amirreza Shaban, Shray Bansal, Zhen Liu, Irfan Essa, and Byron Boots. One-shot learning for semantic segmentation. arXiv preprint arXiv:1709.03410, 2017. 5, 11

    2017 arXiv

  24. [32]

    Indoor segmentation and support inference from rgbd images

    Nathan Silberman, Derek Hoiem, Pushmeet Kohli, and Rob Fergus. Indoor segmentation and support inference from rgbd images. In Computer Vision–ECCV 2012: 12th Eu- ropean Conference on Computer Vision, Florence, Italy, Oc- tober 7-13, 2012, Proceedings, Part V 12 , pages 746–760...

    2012

  25. [33]

    Lamda: Language models for dialog applications

    Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog applications. arXiv preprint arXiv:2201.08239, 2022. 11

    2022 arXiv

  26. [34]

    Diffuse attend and segment: Un- supervised zero-shot segmentation using stable diffusion

    Junjiao Tian, Lavisha Aggarwal, Andrea Colaco, Zsolt Kira, and Mar Gonzalez-Franco. Diffuse attend and segment: Un- supervised zero-shot segmentation using stable diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 3554–3563...

    2024

  27. [35]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth´ee Lacroix, Baptiste Rozi`ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023. 1, 11

    2023 arXiv

  28. [36]

    Plug-and-play diffusion features for text-driven image-to-image translation

    Narek Tumanyan, Michal Geyer, Shai Bagon, and Tali Dekel. Plug-and-play diffusion features for text-driven image-to-image translation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 1921–1930, 2023. 3

    1921

  29. [37]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017. 4

    2017

  30. [38]

    Images speak in images: A generalist painter for in-context visual learning

    Xinlong Wang, Wen Wang, Yue Cao, Chunhua Shen, and Tiejun Huang. Images speak in images: A generalist painter for in-context visual learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6830–6839, 2023. 1, 2, 5, 6, 7, 11, 12, 13, 15

    2023

  31. [39]

    Seggpt: Segmenting ev- erything in context

    Xinlong Wang, Xiaosong Zhang, Yue Cao, Wen Wang, Chunhua Shen, and Tiejun Huang. Seggpt: Segmenting ev- erything in context. arXiv preprint arXiv:2304.03284, 2023. 2, 5, 11

    2023 arXiv

  32. [40]

    Skeleton-in-context: Unified skeleton sequence modeling with in-context learning

    Xinshun Wang, Zhongbin Fang, Xia Li, Xiangtai Li, Chen Chen, and Mengyuan Liu. Skeleton-in-context: Unified skeleton sequence modeling with in-context learning. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2436–2446, 2024. 11

    2024

  33. [41]

    In- context learning unlocked for diffusion models

    Zhendong Wang, Yifan Jiang, Yadong Lu, Pengcheng He, Weizhu Chen, Zhangyang Wang, Mingyuan Zhou, et al. In- context learning unlocked for diffusion models. Advances in Neural Information Processing Systems , 36:8542–8562,

  34. [42]

    Emergent abilities of large language models

    Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, et al. Emergent abilities of large language models. Transactions on Machine Learning Research, 2022. 1, 11

    2022

  35. [43]

    Holistically-nested edge de- tection

    Saining Xie and Zhuowen Tu. Holistically-nested edge de- tection. In Proceedings of the IEEE international conference on computer vision, pages 1395–1403, 2015. 12

    2015

  36. [44]

    Improv: Inpainting-based multimodal prompting for computer vision tasks

    Jiarui Xu, Yossi Gandelsman, Amir Bar, Jianwei Yang, Jian- feng Gao, Trevor Darrell, and Xiaolong Wang. Improv: Inpainting-based multimodal prompting for computer vision tasks. arXiv preprint arXiv:2312.01771, 2023. 1, 2, 5, 6, 7, 11, 12, 13, 18, 19, 20, 21, 22, 23

    2023 arXiv

  37. [45]

    Im- proved distribution matching distillation for fast image syn- thesis

    Tianwei Yin, Micha ¨el Gharbi, Taesung Park, Richard Zhang, Eli Shechtman, Fredo Durand, and William T Freeman. Im- proved distribution matching distillation for fast image syn- thesis. arXiv preprint arXiv:2405.14867, 2024. 17

    2024 arXiv

  38. [46]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR, 2018. 6, 12

    2018

  39. [47]

    What makes good examples for visual in-context learning? Advances in Neural Information Processing Systems, 36:17773–17794,

    Yuanhan Zhang, Kaiyang Zhou, and Ziwei Liu. What makes good examples for visual in-context learning? Advances in Neural Information Processing Systems, 36:17773–17794,

  40. [48]

    Semantic under- standing of scenes through the ade20k dataset

    Bolei Zhou, Hang Zhao, Xavier Puig, Tete Xiao, Sanja Fi- dler, Adela Barriuso, and Antonio Torralba. Semantic under- standing of scenes through the ade20k dataset. International Journal of Computer Vision, 127:302–321, 2019. 12 10 foreground segmentation task, however, in this...

    2019

  41. [50]

    To accommo- date the different spatial scales, we apply Gaussians with smaller variance for facial keypoints, which are relatively finer, and larger variance for body keypoints

    Each heatmap is created by superimposing Gaussian distributions centered on each keypoint. To accommo- date the different spatial scales, we apply Gaussians with smaller variance for facial keypoints, which are relatively finer, and larger variance for body keypoints. These ar...

  42. [51]

    We compute the LPIPS loss and the FID score [16] between the original colored image and the colorized prediction to evaluate the perceptual simi- larity

    for evaluation. We compute the LPIPS loss and the FID score [16] between the original colored image and the colorized prediction to evaluate the perceptual simi- larity. D. Additional Quantitative Results While in Tab. 1 we present the average performance for foreground segmen...

  43. [52]

    LAION5B [30]) that span annotated, unannotated, and sequence images

    and trained on the UVD-V1 [2] dataset, a large-scale vision corpus comprising 50 datasets (e.g. LAION5B [30]) that span annotated, unannotated, and sequence images. Prompt Diffusion is a generative model based on Stable Dif- fusion, jointly finetuned on three forward tasks (i....

  44. [2023]

    1, 5, 6, 7, 11, 12, 13

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.