Pith. sign in

REVIEW 3 major objections 6 minor 85 references

LACONIC: A 3D Layout Adapter for Controllable Image Creation

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A lightweight adapter trained on paired 3D layouts and cameras makes a frozen pretrained text-to-image model respect 3D geometry, support arbitrary viewpoints, and model off-screen scene context.

desk verdict LACONIC is a well-scoped, honest 3D layout adapter with real gains over 2D-conditioned baselines; the 'camera control' claim is narrower than advertised because intrinsics are fixed, but the paper says so itself, and the remaining issues are fixable. read the letter →

arxiv 2507.03257 v2 pith:MXPVNC7K submitted 2025-07-04 cs.CV

Léopold Maillard , Tom Durand , Adrien Ramanana Rahary , Maks Ovsjanikov This is my paper
classification cs.CV
keywords 3Dlayoutconditioningtext-to-imagediffusionadapternetworkcameracontrolsemanticboundingboxessceneeditingmulti-viewconsistencyindoorgeneration
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 tries to show that a small, trainable adapter can give an off-the-shelf text-to-image diffusion model true 3D awareness: given semantic 3D bounding boxes of a room plus a camera pose, the frozen model renders an image that respects object positions, sizes, orientations, and even objects outside the frame. If true, this matters because designers can compose scenes in 3D, render them from any viewpoint, and edit individual objects by moving, resizing, or recaptioning them, while keeping the visual quality of a large pretrained generator. The authors report that on the HyperSim benchmark their adapter reaches FID 9.50, far below the 39.36 of the prior 3D-layout baseline, and that a forced-choice perceptual study favored their images for realism and layout adherence.

What carries the argument

The load-bearing object is the semantic 3D bounding box representation of a scene, combined with a camera reframing transform. Each object is encoded by a shared module that combines sinusoidal positional encodings of position, size, and a continuous rotation representation with text embeddings from the model's own text encoder; an optional floor plan is encoded by a PointNet. A transformer encoder turns the object and floor tokens into a scene conditioning sequence, which is injected into the frozen diffusion backbone through decoupled cross-attention: the image query attends to layout key/values and the result is added to the text-conditioned hidden state with a strength weight $\gamma$. The camera reframing, expressed as $p_i^C = R_C^\top(p_i - p_C)$ and $R_i^C = R_C^\top R_i$, maps world geometry to the target view and lets one scene render consistently from any requested viewpoint.

What would settle it

Train LACONIC on HyperSim-style synthetic scenes but evaluate on a held-out set rendered with a different field of view (for example 60 degrees instead of 90 degrees), keeping layouts and cameras otherwise identical; if FID and SOC degrade sharply relative to the matched-intrinsics setting, the consistent-intrinsics assumption is the load-bearing cause. A more direct check is to generate the same scene at two focal lengths and measure whether projected object centers in the outputs match the ground-truth projections from the camera-reframing equation.

Watch

Extended reading notes

Core claim

LACONIC's central claim is that a lightweight adapter trained on paired (3D layout, camera) to image data can inject 3D awareness into a frozen pretrained text-to-image diffusion model without retraining the backbone. The conditioning signal is an unordered set of objects, each described by a semantic 3D bounding box (center, size, rotation, and an open-vocabulary text caption), plus an optional floor plan point cloud and the camera's extrinsics. Before encoding, every object is transformed into camera coordinates, so the network sees the scene exactly as the target view does; this closed-form reframing is what makes viewpoint control and multi-view consistency possible. Because every object in the 3D scene is part of the conditioning sequence, off-screen items influence the render, for example letting a window outside the frame light the room. At inference the adapter supports camera control, per-object geometric edits, open-vocabulary restyling, and scene-level text prompting, and the paper reports that it outperforms the prior supervised baseline and a from-scratch-trained diffusion model on FID, KID, IS, and its object-level SOC metric.

Load-bearing premise

Every training and test image must be rendered with the same camera intrinsics; if a dataset mixes different focal lengths or real photos, the camera-reframing transform misplaces every object in the image, breaking both layout adherence and viewpoint control.

Editorial extensions

If this is right

  • A single trained adapter can render the same 3D scene from arbitrary camera viewpoints with consistent object structure, because the conditioning is view-independent.
  • Users can move, rotate, resize, add, or remove individual objects in 3D and regenerate the image; global text restyling changes materials and style on the relevant objects without leaking to floors or ceilings.
  • Because the full 3D context is conditioned, off-screen elements are represented, so deleting a window changes the global illumination of the generated room.
  • The adapter transfers to a different diffusion backbone architecture with only a retrained adapter, suggesting the 3D layout control is not tied to a specific UNet implementation.
  • Object-level adherence can be measured with the paper's SOC metric, which crops visible objects and correlates their CLIP similarity with their captions.

Reading between the lines

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

  • A natural stress test is to train or evaluate on real photographs with unknown or varied camera intrinsics; the paper's own limitation section suggests this is where the camera-reframing assumption would break, so a variant that estimates or marginalizes over intrinsics is the obvious next step.
  • The same conditioning could be chained with an inverse step: use a depth or layout estimator on a single image to recover a semantic 3D layout, then use LACONIC to re-render from new viewpoints, turning the method into a 3D-aware image-to-video or editing pipeline.
  • Because the scene sequence treats objects as unordered tokens, the adapter could likely be trained jointly with a 3D scene synthesis model so a generated layout can be rendered immediately; the authors note NeRF-style continuous scene representations as future work.
  • The consistent-camera-intrinsics assumption suggests a testable extension: learn or predict a per-sample focal length and feed it into the reframing transform, broadening applicability beyond synthetic indoor renders.
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

3 major / 6 minor

Summary. LACONIC proposes a trainable adapter that conditions a frozen text-to-image diffusion model on an explicit 3D semantic layout: per-object 3D bounding boxes with free-form language captions, an optional floor plan point cloud, and a target camera pose. The adapter encodes objects with a shared MLP and a transformer encoder, then injects the resulting tokens through decoupled cross-attention (Eqs. 2-3), with supervised training on rendered layout-image pairs under the assumption of consistent camera intrinsics. The paper reports strong quantitative gains over SceneCraft and a from-scratch diffusion baseline on HyperSim (Table 1), an ablation of the camera reframing and transformer encoder (Table 2), a perceptual study favoring LACONIC, and qualitative demonstrations of multi-view generation, object editing, text restyling, and DiT compatibility. The central claim is that semantic 3D boxes plus camera pose are a sufficient, view-consistent control signal for high-quality image synthesis from a frozen prior.

Significance. Assuming the results hold, LACONIC is a significant practical contribution to layout-guided image synthesis. It combines a parametric 3D representation with a lightweight adapter, and the evidence is consistent: large FID/KID margins over SceneCraft (9.50 vs 39.36 FID), ablations isolating camera reframing and the transformer encoder, a 638-vote perceptual study favoring LACONIC by 71.2% on realism and 89.0% on layout adherence, and a DiT-backbone demonstration. The supervised objective is standard and not circular. The main caveat is that the camera-control claim is restricted by the fixed-intrinsics assumption, which is acknowledged in Section D.

major comments (3)
  1. [Section 3.2, Eq. (4), Section D] The camera-conditioning mechanism is invariant to intrinsic parameters, and this limits the headline "camera control" claim. Eq. (4) feeds the adapter only the extrinsic mapping (p_C, R_C); focal length and principal point are not part of y. Under the stated assumption of consistent intrinsics the model can memorize the single projection, but at test time a user-provided camera with different intrinsics maps the same camera-space tokens to different pixel locations, so the layout condition is systematically misaligned. Section D explicitly attributes "unnatural perspective... distorted floors and objects" to this assumption. Please either incorporate intrinsics into the conditioning (or train with varying intrinsics), or qualify the abstract and Section 4 claims to "camera control with fixed intrinsic parameters" and report results under that restriction.
  2. [Section A.2.1 and Abstract] The claim to "account for the entire context of a scene, i.e., both on and off-screen items" is not supported by the training/evaluation protocol on HyperSim. The conditioning sequence is built by prioritizing visible objects from the target rendering and is capped at N=50, while HyperSim scenes have a median of 54 objects (and average 121), so off-screen objects will frequently be dropped. The only supporting evidence is the qualitative window-removal example in Figure 4. Please provide a controlled experiment that varies off-screen objects (e.g., ablating off-screen lighting sources) and quantifies their effect, or restrict the claim to the cases actually evaluated.
  3. [Section 4.2 and Figure 1(e-h)] Multi-view consistency is presented as a core capability ("maintaining a consistent 3D structure across views"), but no quantitative evaluation of cross-view consistency is provided. The three views in Figure 1 are qualitative, and the evaluation in Section 4.1 does not include a correspondence, depth, or pose-based consistency metric. Please add a quantitative multi-view consistency evaluation, for example by generating images from held-out camera poses for the same 3D layout and measuring geometric correspondence against ground-truth renderings, or by a user study specifically targeting view consistency.
minor comments (6)
  1. [Table 1] "w/otext prompt" is a typo; it should read "w/o text prompt".
  2. [Section 4.1 vs Supplementary A.2.1] The main paper states 326 unique HyperSim layouts, while the supplementary material states 323 unique scenes; please reconcile the numbers.
  3. [Supplementary A.6] The SOC metric thresholds (alpha=0.4, 2% area filter, exclusion of "other" NYU40 class) are not accompanied by a sensitivity analysis; a short robustness check would help assess the metric's stability.
  4. [Supplementary C.3] The perceptual study reports preference rates of 71.2% and 89.0% but no confidence intervals or statistical significance test; a binomial test or bootstrapped CI would strengthen the claim.
  5. [Supplementary B] The DiT generalization section is qualitative only; consider adding a sentence noting that no quantitative metrics are reported for the SD3 adapter, or adding a small quantitative comparison.
  6. [Figure 13] The caption labels (a)-(c) are not visually marked in the figure itself, making the failure examples harder to map; please add the labels directly in the image.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: LACONIC's layout-conditioned generation is a supervised adapter trained against an external image target, and its headline results come from independent image-quality metrics and an external baseline.

full rationale

The paper's derivation chain is a standard supervised adapter training loop: Eq. (1) optimizes a denoising objective conditioned on a 3D layout y drawn from paired (layout, camera, image) data; Eq. (4) expresses object poses in camera coordinates; and the adapter is trained on HyperSim and a custom bedroom dataset. The layout input is external conditioning, not a fitted constant, and the evaluated outputs (FID/KID/IS/CLIP/SOC) compare generated images against external reference distributions or pretrained CLIP embeddings. The SOC metric is newly proposed and uses ground-truth 2D boxes for crop selection and CLIP for semantic matching; its filtering choices (alpha = 0.4, 2% area floor, NYU40 exclusion) are evaluation-design decisions and do not make the reported table values true by construction. The assumption of consistent camera intrinsics (Sec. 3.2) and the admitted perspective failures (Sec. D) bound the claimed camera control rather than reduce the method to its inputs; Eq. (4) is a geometric transformation, not a fitted re-prediction of the output. Self-citations ([38] Debara, with [46, 67, 74]) are invoked only as architectural inspiration for the layout encoder and are not load-bearing as evidence for the central claim. No step in the derivation is equivalent to its input by definition.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The paper postulates no new physical or model entities. Its contributions are representational (camera-reframed 3D box tokens) and architectural (adapter), both implemented with existing components. The SOC metric is a new measurement protocol rather than an inferred entity. The load-bearing assumptions are data-domain premises about camera intrinsics, layout sufficiency, VLM caption reliability, and CLIP as ground truth; all are standard for the subfield but unverified in this paper.

free parameters (4)
  • gamma (adapter strength, Eq. 3) = 0.0 to 0.85 at inference, user-set
    Scales the layout condition, H = H^c + gamma * H^y (Eq. 3). Figure 6 shows outputs transitioning from text-driven to layout-driven as gamma grows, so any claim about 'adherence to both conditions' depends on this hand-set knob.
  • SOC visibility threshold alpha = 0.4
    Objects whose projected 3D box area ratio is below 0.4 are dropped from the SOC evaluation set, along with tiny objects under 2% image area and non-descriptive NYU40 labels (Section A.6). These author-chosen thresholds define the new object-level metric used in the headline table.
  • Layout dropout rate pdrop = 0.15
    Classifier-free guidance drop probability for the 3D layout condition during training (Section A.3). Standard practice, but it interacts with the reported layout-adherence behavior.
  • Max conditioning objects N=50 = 50
    HyperSim scenes average 121 objects but are truncated to 50 tokens, prioritizing visible objects via instance maps (Section A.2.1). This truncation is a hidden factor in layout-adherence performance and weakens the 'entire context' claim.
assumptions (5)
  • domain assumption Consistent camera intrinsics across all training samples (Section 3.2).
    The paper states 'we assume that the image samples x from the data distribution are rendered using consistent camera intrinsics.' Section D later attributes distorted-perspective failures to this assumption, so it is load-bearing for viewpoint control.
  • domain assumption Semantic 3D bounding boxes plus captions and camera pose are a sufficient generative condition for the target image.
    The supervised pairs (S, C) to x in Eq. (1) treat boxes and captions as the complete scene description; object geometry, materials, lighting, and style must be supplied by the pretrained prior. Section 5 concedes the model stays within its training-domain prior.
  • domain assumption VLM-generated object captions are accurate enough to supervise object appearance.
    The 72,000-layout custom dataset uses LLaVA captions from a fixed instruction with no human verification (Section A.2.2). Caption noise transfers directly into the learned object semantics.
  • domain assumption CLIP similarity between an object crop and its caption measures layout and semantic adherence.
    The SOC metric assumes CLIP embeddings capture both placement (via cropping from the projected box) and semantics. The paper does not validate SOC against human object-level judgments.
  • standard math Standard background: diffusion denoising objective, classifier-free guidance, attention mechanisms.
    Equation (1) and Eqs. (2)-(3) rely on published DDPM, CFG, and cross-attention theory, cited to [18, 19, 55, 69, 79].

how reviews work

0 comments
Cite this review

Pith. "Pith review of LACONIC: A 3D Layout Adapter for Controllable Image Creation." pith.science (2026). https://pith.science/paper/MXPVNC7K

@misc{pith2026250703257,
  author       = {Pith},
  title        = {Pith review of: LACONIC: A 3D Layout Adapter for Controllable Image Creation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MXPVNC7K}},
  note         = {Machine review of arXiv:2507.03257}
}
read the original abstract

Existing generative approaches for guided image synthesis of multi-object scenes typically rely on 2D controls in the image or text space. As a result, these methods struggle to maintain and respect consistent three-dimensional geometric structure, underlying the scene. In this paper, we propose a novel conditioning approach, training method and adapter network that can be plugged into pretrained text-to-image diffusion models. Our approach provides a way to endow such models with 3D-awareness, while leveraging their rich prior knowledge. Our method supports camera control, conditioning on explicit 3D geometries and, for the first time, accounts for the entire context of a scene, i.e., both on and off-screen items, to synthesize plausible and semantically rich images. Despite its multi-modal nature, our model is lightweight, requires a reasonable number of data for supervised learning and shows remarkable generalization power. We also introduce methods for intuitive and consistent image editing and restyling, e.g., by positioning, rotating or resizing individual objects in a scene. Our method integrates well within various image creation workflows and enables a richer set of applications compared to previous approaches.

Figures

Figures reproduced from arXiv: 2507.03257 by the authors.

Figure 1
Figure 1. Overview of LACONIC Capabilities and Applications. Our model generates realistic renderings from an input semantic 3D layout and target viewpoint (a), while leveraging the comprehensive knowledge of a text-to-image prior (b–d). A given 3D scene can be rendered from multiple camera poses (e) while maintaining a consistent 3D structure across views (f–h). Finally, objects can be individually manipulated (i) by editing… view at source ↗
Figure 2
Figure 2. Comparison of high-level conditioning input representations for describing a 3D scene. Relying solely on text descriptions (a) can make it difficult to convey complex spatial relations. Conditioning via 2D bounding boxes (b) can lead to ambiguity in perspective and does not account for out-of-bound objects. Semantic and depth maps rendered from 3D bounding boxes (c), as introduced in recent work [77], cannot always … view at source ↗
Figure 3
Figure 3. LACONIC Architecture & Pipeline Overview. From an input 3D layout \protect \mathcal {S} and camera pose \protect \mathcal {C}, trainable modules embeds geometric and semantic properties of individual objects to guide a text-to-image diffusion prior in denoising a target rendering. Camera control is enabled by expressing spatial features (p,R) from the input object 3D bounding boxes \protect \mathcal {O} in the coord… view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Iterative Scene Editing Results. From left to right: given an input semantic 3D layout and camera viewpoint (a), we render the scene both without (b) and with (c) global text prompt conditioning. Then, individual objects are subsequently moved (d), resized (e) and re-c…
Figure 5
Figure 5. Figure 5: 3D layout-guided image synthesis baseline comparisons. Our method produces more detailed and natural images compared to baseline approaches. Methods leveraging our 3D layout encoder (DM-FS & ours) better represent the guiding layout, while our adapter￾based approach ad…
Figure 7
Figure 7. Figure 7: Object selection for the SOC metric. Obstructed and out-of-bounds objects are identified from their 2D and 3D bound￾ing box annotations and filtered out from the evaluation set. Our metric takes as input a synthesized image, a guiding 3D semantic layout and 2D object b…
Figure 6
Figure 6. Figure 6: Impact of the adapter strength on generated images. From a text caption c= “A Retro Synthwave Indoor Bedroom”, adherence with the input 3D layout improves accordingly with the parameter \gamma from Equation (3), achieving strict control at higher scales. We can observe…
Figure 8
Figure 8. Figure 8: Additional layout-guided image synthesis results from diverse prompts. The generated images (Right) from the input 3D layouts (Left) and text prompts demonstrate our method’s strong adherence to both conditions. Notably, LACONIC produces high-quality results across var…
Figure 9
Figure 9. Figure 9: Text-driven DiT synthesis results. Given an input 3D layout, viewpoint and caption (a), LACONIC with Stable Diffu￾sion 3 [8] supports adjusting the adapter strength to balance fidelity to the text prompt against adherence to the input layout (b–d). A.4. Inference Setti…
Figure 10
Figure 10. Figure 10: Layout-guided image synthesis results with a DiT-based backbone. Our LACONIC adapter successfully conditions Stable Diffusion 3 [8], demonstrating compatibility with modern DiT architectures. The generated images (Right) show strong adherence to the input 3D layouts a…
Figure 11
Figure 11. Figure 11: Qualitative comparison with Build-A-Scene [7] given a common 3D layout and viewpoint. For each method, from left to right: input scene representation, generation result for prompt c_1= “a cozy bedroom with a wooden floor”, and for c_2= “a Van-Gogh style bedroom”. Buil…
Figure 12
Figure 12. Figure 12: Conditioning inputs for the SceneCraft [77] base￾line. The model is conditioned on depth (c) and segmentation (d) maps, which are rendered from the 3D bounding box layout (a) underlying a ground-truth bedroom image (b). C.4. Qualitative Results We provide additional q…
Figure 13
Figure 13. Figure 13: illustrates failure cases and known limitations. Although our method demonstrates state-of-the-art 3D lay￾out adherence, it can occasionally generate results with in￾consistencies, such as missing objects, visual artifacts, or distorted perspectives. E. Societal Impac…
Figure 14
Figure 14. Figure 14: Perceptual Study Interface. Users are prompted to independently select which generation result is (i) the most realistic and (ii) more in line with the input 3D layout and viewpoint. (a) Input 3D Layout (b) Ground-Truth Image (c) SceneCraft (c) Ours [PITH_FULL_IMAGE:…
Figure 15
Figure 15. Figure 15: Comparison with SceneCraft [77] on 3D layout-guided image synthesis. Our method demonstrates superior realism and adherence to the input 3D layout and viewpoint on our custom bedroom dataset. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_15.png]
Figure 16
Figure 16. Figure 16: Additional 3D layout-guided image synthesis baseline comparisons (1/2). We can observe that our method produces more natural images that better respect the input 3D layout. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_16.png]
Figure 17
Figure 17. Figure 17: Additional 3D layout-guided image synthesis baseline comparisons (2/2). 21 [PITH_FULL_IMAGE:figures/full_fig_p021_17.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

85 extracted references · 63 canonical work pages

  1. [1]

    Layer normalization

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hin- ton. Layer normalization. arXiv preprint arXiv:1607.06450,

    arXiv

  2. [2]

    Loosec- ontrol: Lifting controlnet for generalized depth conditioning

    Shariq Farooq Bhat, Niloy Mitra, and Peter Wonka. Loosec- ontrol: Lifting controlnet for generalized depth conditioning. In ACM SIGGRAPH 2024 Conference Papers , pages 1–11,

    2024

  3. [3]

    Sutherland, Michael Arbel, and Arthur Gretton

    Mikołaj Bi ´nkowski, Dougal J. Sutherland, Michael Arbel, and Arthur Gretton. Demystifying MMD GANs. In Inter- national Conference on Learning Representations (ICLR) ,

  4. [4]

    Generative novel view synthesis with 3d-aware diffusion models

    Eric R Chan, Koki Nagano, Matthew A Chan, Alexander W Bergman, Jeong Joon Park, Axel Levy, Miika Aittala, Shalini De Mello, Tero Karras, and Gordon Wetzstein. Generative novel view synthesis with 3d-aware diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4217–4229, 2023. 2

    2023

  5. [5]

    Training-free layout control with cross-attention guidance

    Minghao Chen, Iro Laina, and Andrea Vedaldi. Training-free layout control with cross-attention guidance. In Proceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 5343–5353, 2024. 3

    2024

  6. [6]

    Set-the-scene: Global-local training for generating controllable nerf scenes

    Dana Cohen-Bar, Elad Richardson, Gal Metzer, Raja Giryes, and Daniel Cohen-Or. Set-the-scene: Global-local training for generating controllable nerf scenes. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 2920–2929, 2023. 3

    2023

  7. [7]

    Build-a-scene: Interactive 3d layout control for diffusion-based image gen- eration

    Abdelrahman Eldesokey and Peter Wonka. Build-a-scene: Interactive 3d layout control for diffusion-based image gen- eration. In The Thirteenth International Conference on Learning Representations (ICLR), 2025. 2, 3, 8, 17

    work page 2025

  8. [8]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. In Forty-first International Conference on Machine Learning (ICML), 2024. 8, 15, 16

    work page 2024

Show all 85 references
  1. [9]

    PyTorch Lightning, 2019

    William Falcon and The PyTorch Lightning team. PyTorch Lightning, 2019. 14

    2019

  2. [10]

    Ctrl-room: Controllable text-to-3d room meshes generation with layout constraints

    Chuan Fang, Yuan Dong, Kunming Luo, Xiaotao Hu, Rakesh Shrestha, and Ping Tan. Ctrl-room: Controllable text-to-3d room meshes generation with layout constraints. arXiv preprint arXiv:2310.03602, 2023. 2, 3

    2023

  3. [11]

    The stable signature: Rooting watermarks in latent diffusion models

    Pierre Fernandez, Guillaume Couairon, Herv ´e J ´egou, Matthijs Douze, and Teddy Furon. The stable signature: Rooting watermarks in latent diffusion models. In Proceed- ings of the IEEE/CVF International Conference on Com- puter Vision (ICCV), pages 22466–22477, 2023. 18

    2023

  4. [12]

    3d-front: 3d furnished rooms with layouts and semantics

    Huan Fu, Bowen Cai, Lin Gao, Ling-Xiao Zhang, Jiaming Wang, Cao Li, Qixun Zeng, Chengyue Sun, Rongfei Jia, Binqiang Zhao, et al. 3d-front: 3d furnished rooms with layouts and semantics. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV) , pages...

    2021

  5. [13]

    Srinivasan, Jonathan T

    Ruiqi Gao, Aleksander Holynski, Philipp Henzler, Arthur Brussee, Ricardo Martin Brualla, Pratul P. Srinivasan, Jonathan T. Barron, and Ben Poole. CAT3d: Create anything in 3d with multi-view diffusion models. InThe Thirty-eighth Annual Conference on Neural Information Processi...

    2024

  6. [14]

    Nerfdiff: Single-image view synthesis with nerf-guided dis- tillation from 3d-aware diffusion

    Jiatao Gu, Alex Trevithick, Kai-En Lin, Joshua M Susskind, Christian Theobalt, Lingjie Liu, and Ravi Ramamoorthi. Nerfdiff: Single-image view synthesis with nerf-guided dis- tillation from 3d-aware diffusion. In International Confer- ence on Machine Learning (ICML) , pages 118...

    2023

  7. [15]

    Gaussian error linear units (gelus)

    Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415, 2016. 14

    2016 arXiv

  8. [16]

    Clipscore: A reference-free evaluation met- ric for image captioning

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. Clipscore: A reference-free evaluation met- ric for image captioning. arXiv preprint arXiv:2104.08718,

    arXiv

  9. [17]

    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 (NeurIPS), 30, 2017. 7

    2017

  10. [18]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. In NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications, 2021. 5

    2021

  11. [19]

    Denoising diffu- sion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. Advances in Neural Information Processing Systems (NeurIPS), 33:6840–6851, 2020. 2, 14

    2020

  12. [20]

    Video dif- fusion models

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video dif- fusion models. Advances in Neural Information Processing Systems (NeurIPS), 35:8633–8646, 2022. 2

    2022

  13. [21]

    Text2room: Extracting textured 3d meshes from 2d text-to-image models

    Lukas H ¨ollein, Ang Cao, Andrew Owens, Justin Johnson, and Matthias Nießner. Text2room: Extracting textured 3d meshes from 2d text-to-image models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 7909–7920, 2023. 2

    2023

  14. [22]

    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. InIn- ternational Conference on Learning Representations (ICLR),

  15. [23]

    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. Advances in Neural Information Processing Systems (NeurIPS), 35:26565–26577, 2022. 2

    2022

  16. [24]

    Controllable text-to-image generation

    Bowen Li, Xiaojuan Qi, Thomas Lukasiewicz, and Philip Torr. Controllable text-to-image generation. Advances in 9 neural information processing systems (NeurIPS), 32, 2019. 2

    2019

  17. [25]

    Blip: Bootstrapping language-image pre-training for uni- fied vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for uni- fied vision-language understanding and generation. In In- ternational Conference on Machine Learning (ICML), pages 12888–12900. PMLR, 2022. 5

    2022

  18. [26]

    Grains: Generative re- cursive autoencoders for indoor scenes

    Manyi Li, Akshay Gadi Patil, Kai Xu, Siddhartha Chaudhuri, Owais Khan, Ariel Shamir, Changhe Tu, Baoquan Chen, Daniel Cohen-Or, and Hao Zhang. Grains: Generative re- cursive autoencoders for indoor scenes. ACM Transactions on Graphics (TOG), 38(2):1–16, 2019. 3

    2019

  19. [27]

    Gligen: Open-set grounded text-to-image generation

    Yuheng Li, Haotian Liu, Qingyang Wu, Fangzhou Mu, Jian- wei Yang, Jianfeng Gao, Chunyuan Li, and Yong Jae Lee. Gligen: Open-set grounded text-to-image generation. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 22511–22521, 2...

    2023

  20. [28]

    Unraveling the hidden environmental impacts of ai solutions for environment

    Anne-Laure Ligozat, Julien Lef `evre, Aur ´elie Bugeau, and Jacques Combaz. Unraveling the hidden environmental impacts of ai solutions for environment. arXiv preprint arXiv:2110.11822, 2021. 18

    2021 arXiv

  21. [29]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maxim- ilian Nickel, and Matthew Le. Flow matching for genera- tive modeling. In The Eleventh International Conference on Learning Representations (ICLR), 2023. 2, 16

    2023

  22. [30]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. Advances in Neural Information Processing Systems (NeurIPS) , 36:34892–34916, 2023. 5, 14

    2023

  23. [31]

    SINGAPO: Single image con- trolled generation of articulated parts in objects

    Jiayi Liu, Denys Iliash, Angel X Chang, Manolis Savva, and Ali Mahdavi-Amiri. SINGAPO: Single image con- trolled generation of articulated parts in objects. In The Thirteenth International Conference on Learning Represen- tations (ICLR), 2025. 2

    2025

  24. [32]

    One-2-3-45: Any single image to 3d mesh in 45 seconds without per-shape optimiza- tion

    Minghua Liu, Chao Xu, Haian Jin, Linghao Chen, Mukund Varma T, Zexiang Xu, and Hao Su. One-2-3-45: Any single image to 3d mesh in 45 seconds without per-shape optimiza- tion. Advances in Neural Information Processing Systems (NeurIPS), 36, 2023. 2

    2023

  25. [33]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 14

    2017 arXiv

  26. [34]

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

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for dif- fusion probabilistic model sampling in around 10 steps. Ad- vances in Neural Information Processing Systems (NeurIPS), 35:5775–5787, 2022. 15

    2022

  27. [35]

    Power hungry processing: Watts driving the cost of ai deployment? In Proceedings of the 2024 ACM conference on fairness, ac- countability, and transparency, pages 85–99, 2024

    Sasha Luccioni, Yacine Jernite, and Emma Strubell. Power hungry processing: Watts driving the cost of ai deployment? In Proceedings of the 2024 ACM conference on fairness, ac- countability, and transparency, pages 85–99, 2024. 18

    2024

  28. [36]

    Repaint: Inpainting using denoising diffusion probabilistic models

    Andreas Lugmayr, Martin Danelljan, Andres Romero, Fisher Yu, Radu Timofte, and Luc Van Gool. Repaint: Inpainting using denoising diffusion probabilistic models. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11461–11471, 2022. 6

    2022

  29. [37]

    Gonzalez, Zhifeng Chen, Russ Salakhutdi- nov, and Ion Stoica

    Michael Luo, Justin Wong, Brandon Trabucco, Yanping Huang, Joseph E. Gonzalez, Zhifeng Chen, Russ Salakhutdi- nov, and Ion Stoica. Stylus: Automatic adapter selection for diffusion models. In The Thirty-eighth Annual Conference on Neural Information Processing Systems (NeurIPS...

    2024

  30. [38]

    Debara: Denoising-based 3d room arrangement generation

    L ´eopold Maillard, Nicolas Sereyjol-Garros, Tom Durand, and Maks Ovsjanikov. Debara: Denoising-based 3d room arrangement generation. In Advances in Neural Informa- tion Processing Systems (NeurIPS) , pages 109202–109232,

  31. [39]

    Object 3dit: Language-guided 3d-aware image editing

    Oscar Michel, Anand Bhattad, Eli VanderBilt, Ranjay Kr- ishna, Aniruddha Kembhavi, and Tanmay Gupta. Object 3dit: Language-guided 3d-aware image editing. Advances in Neural Information Processing Systems (NeurIPS), 36, 2023. 2

    2023

  32. [40]

    T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models

    Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4296–4304, 2024. 2, 3

    2024

  33. [41]

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

    Alexander Quinn Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob Mcgrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealis- tic image generation and editing with text-guided diffusion models. In International Conference on Machine Learning (I...

    2022

  34. [42]

    Maxime Oquab, Timoth ´ee Darcet, Th´eo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael ...

    2024

  35. [43]

    Diffusion handles enabling 3d edits for diffusion models by lifting ac- tivations to 3d

    Karran Pandey, Paul Guerrero, Matheus Gadelha, Yannick Hold-Geoffroy, Karan Singh, and Niloy J Mitra. Diffusion handles enabling 3d edits for diffusion models by lifting ac- tivations to 3d. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (...

    2024

  36. [44]

    Generative layout modeling using con- straint graphs

    Wamiq Para, Paul Guerrero, Tom Kelly, Leonidas J Guibas, and Peter Wonka. Generative layout modeling using con- straint graphs. In Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision (ICCV), pages 6690– 6700, 2021. 3

    2021

  37. [45]

    Benchmark for compositional text-to- image synthesis

    Dong Huk Park, Samaneh Azadi, Xihui Liu, Trevor Darrell, and Anna Rohrbach. Benchmark for compositional text-to- image synthesis. In Thirty-fifth Conference on Neural Infor- mation Processing Systems Datasets and Benchmarks Track (Round 1), 2021. 2

    2021

  38. [46]

    Atiss: Autore- gressive transformers for indoor scene synthesis

    Despoina Paschalidou, Amlan Kar, Maria Shugrina, Karsten Kreis, Andreas Geiger, and Sanja Fidler. Atiss: Autore- gressive transformers for indoor scene synthesis. Advances in Neural Information Processing Systems (NeurIPS) , 34: 12013–12026, 2021. 3, 4, 5, 13 10

    2021

  39. [47]

    Pytorch: An im- perative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An im- perative style, high-performance deep learning library. Ad- vances in Neural Information Processing Systems ...

    2019

  40. [48]

    Scalable diffusion mod- els with transformers

    William Peebles and Saining Xie. Scalable diffusion mod- els with transformers. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV) , pages 4195–4205, 2023. 2, 16

    2023

  41. [49]

    Grounded text-to-image synthesis with attention refocusing

    Quynh Phung, Songwei Ge, and Jia-Bin Huang. Grounded text-to-image synthesis with attention refocusing. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognitio (CVPR), pages 7932–7942, 2024. 2

    2024

  42. [50]

    Movie gen: A cast of media foundation models

    Adam Polyak, Amit Zohar, Andrew Brown, Andros Tjandra, Animesh Sinha, Ann Lee, Apoorv Vyas, Bowen Shi, Chih- Yao Ma, Ching-Yao Chuang, et al. Movie gen: A cast of media foundation models. arXiv preprint arXiv:2410.13720,

    arXiv

  43. [51]

    Barron, and Ben Milden- hall

    Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion. In The Eleventh International Conference on Learning Representa- tions (ICLR), 2023. 3

    2023

  44. [52]

    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 Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 652–660, 2017. 5, 13

    2017

  45. [53]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In International Conference on Machine Learning...

    2021

  46. [54]

    Hypersim: A photorealistic syn- thetic dataset for holistic indoor scene understanding

    Mike Roberts, Jason Ramapuram, Anurag Ranjan, Atulit Kumar, Miguel Angel Bautista, Nathan Paczan, Russ Webb, and Joshua M Susskind. Hypersim: A photorealistic syn- thetic dataset for holistic indoor scene understanding. In Proceedings of the IEEE/CVF International Conference o...

    2021

  47. [55]

    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 (CVPR), pages 10684–10695, 2022. 2, 3, 7, 13, 15

    2022

  48. [56]

    U- net: Convolutional networks for biomedical image segmen- tation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- net: Convolutional networks for biomedical image segmen- tation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, pa...

    2015

  49. [57]

    Ipadapter- instruct: Resolving ambiguity in image-based conditioning using instruct prompts

    Ciara Rowles, Shimon Vainer, Dante De Nigris, Slava Elizarov, Konstantin Kutsy, and Simon Donn ´e. Ipadapter- instruct: Resolving ambiguity in image-based conditioning using instruct prompts. In European Conference on Com- puter Vision (ECCV), pages 54–70. Springer, 2025. 16

    2025

  50. [58]

    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. Advances in Neural Information...

    2022

  51. [59]

    Improved techniques for training gans

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans. Advances in Neural Information Process- ing Systems (NeurIPS), 29, 2016. 7

    2016

  52. [60]

    Watermark anything with localized messages

    Tom Sander, Pierre Fernandez, Alain Durmus, Teddy Furon, and Matthijs Douze. Watermark anything with localized messages. arXiv preprint arXiv:2411.07231, 2024. 18

    2024 arXiv

  53. [61]

    Controlroom3d: Room gen- eration using semantic proxy rooms

    Jonas Schult, Sam Tsai, Lukas H ¨ollein, Bichen Wu, Jialiang Wang, Chih-Yao Ma, Kunpeng Li, Xiaofang Wang, Felix Wimbauer, Zijian He, et al. Controlroom3d: Room gen- eration using semantic proxy rooms. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

    2024

  54. [62]

    MVDream: Multi-view diffusion for 3d gen- eration

    Yichun Shi, Peng Wang, Jianglong Ye, Long Mai, Kejie Li, and Xiao Yang. MVDream: Multi-view diffusion for 3d gen- eration. In The Twelfth International Conference on Learn- ing Representations, 2024. 3

    2024

  55. [63]

    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

  56. [64]

    Make-a-video: Text-to-video generation without text-video data

    Uriel Singer, Adam Polyak, Thomas Hayes, Xi Yin, Jie An, Songyang Zhang, Qiyuan Hu, Harry Yang, Oron Ashual, Oran Gafni, Devi Parikh, Sonal Gupta, and Yaniv Taigman. Make-a-video: Text-to-video generation without text-video data. In The Eleventh International Conference on Lea...

    2023

  57. [65]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020. 15

    2010 arXiv

  58. [66]

    Score-based generative modeling through stochastic differential equa- tions

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. In International Conference on Learning Represen- tations (ICLR), 2021. 2

    2021

  59. [67]

    Diffuscene: Denoising diffu- sion models for generative indoor scene synthesis

    Jiapeng Tang, Yinyu Nie, Lev Markhasin, Angela Dai, Justus Thies, and Matthias Nießner. Diffuscene: Denoising diffu- sion models for generative indoor scene synthesis. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 20507–20...

    2024

  60. [68]

    Mvdiffusion: Enabling holistic multi- view image generation with correspondence-aware diffu- sion

    Shitao Tang, Fuyang Zhang, Jiacheng Chen, Peng Wang, and Yasutaka Furukawa. Mvdiffusion: Enabling holistic multi- view image generation with correspondence-aware diffu- sion. In Advances in Neural Information Processing Systems (NeurIPS), pages 51202–51233, 2023. 2

    2023

  61. [69]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in Neural 11 Information Processing Systems (NeurIPS) , 30, 2017. 3, 4, 5, 13, 14

    2017

  62. [70]

    Diffusers: State-of-the-art diffu- sion models

    Patrick von Platen, Suraj Patil, Anton Lozhkov, Pedro Cuenca, Nathan Lambert, Kashif Rasul, Mishig Davaadorj, Dhruv Nair, Sayak Paul, William Berman, Yiyi Xu, Steven Liu, and Thomas Wolf. Diffusers: State-of-the-art diffu- sion models. https://github.com/huggingface/ diffusers...

    2022

  63. [71]

    Diffusion models are geometry critics: Single im- age 3d editing using pre-trained diffusion priors

    Ruicheng Wang, Jianfeng Xiang, Jiaolong Yang, and Xin Tong. Diffusion models are geometry critics: Single im- age 3d editing using pre-trained diffusion priors. In Euro- pean Conference on Computer Vision (ECCV) , pages 441–

  64. [72]

    Seman- tic image synthesis via diffusion models

    Weilun Wang, Jianmin Bao, Wengang Zhou, Dongdong Chen, Dong Chen, Lu Yuan, and Houqiang Li. Seman- tic image synthesis via diffusion models. arXiv preprint arXiv:2207.00050, 2022. 3

    2022

  65. [73]

    Novel view synthesis with diffusion models

    Daniel Watson, William Chan, Ricardo Martin Bru- alla, Jonathan Ho, Andrea Tagliasacchi, and Mohammad Norouzi. Novel view synthesis with diffusion models. In The Eleventh International Conference on Learning Repre- sentations (ICLR), 2023. 2

    2023

  66. [74]

    Lego-net: Learning regular rearrangements of ob- jects in rooms

    Qiuhong Anna Wei, Sijie Ding, Jeong Joon Park, Rahul Sajnani, Adrien Poulenard, Srinath Sridhar, and Leonidas Guibas. Lego-net: Learning regular rearrangements of ob- jects in rooms. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), p...

    2023

  67. [75]

    Hud- son, Igor Gilitschenski, Yusuf Aytar, Sjoerd van Steenkiste, Kelsey R Allen, and Thomas Kipf

    Ziyi Wu, Yulia Rubanova, Rishabh Kabra, Drew A. Hud- son, Igor Gilitschenski, Yusuf Aytar, Sjoerd van Steenkiste, Kelsey R Allen, and Thomas Kipf. Neural assets: 3d-aware multi-object scene synthesis with image diffusion models. In The Thirty-eighth Annual Conference on Neural...

    2024

  68. [76]

    3d-aware image generation using 2d diffusion mod- els

    Jianfeng Xiang, Jiaolong Yang, Binbin Huang, and Xin Tong. 3d-aware image generation using 2d diffusion mod- els. In Proceedings of the IEEE/CVF International Confer- ence on Computer Vision (ICCV) , pages 2383–2393, 2023. 2

    2023

  69. [77]

    Scenecraft: Layout-guided 3d scene generation

    Xiuyu Yang, Yunze Man, Junkun Chen, and Yu-Xiong Wang. Scenecraft: Layout-guided 3d scene generation. Ad- vances in Neural Information Processing Systems (NeurIPS), 37:82060–82084, 2024. 2, 3, 5, 6, 7, 8, 14, 15, 17, 18, 19

    2024

  70. [78]

    Reco: Region-controlled text-to-image genera- tion

    Zhengyuan Yang, Jianfeng Wang, Zhe Gan, Linjie Li, Kevin Lin, Chenfei Wu, Nan Duan, Zicheng Liu, Ce Liu, Michael Zeng, et al. Reco: Region-controlled text-to-image genera- tion. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), page...

    2023

  71. [79]

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

    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

  72. [80]

    Scannet++: A high-fidelity dataset of 3d in- door scenes

    Chandan Yeshwanth, Yueh-Cheng Liu, Matthias Nießner, and Angela Dai. Scannet++: A high-fidelity dataset of 3d in- door scenes. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 12–22, 2023. 2

    2023

  73. [81]

    Commonscenes: Generating commonsense 3d indoor scenes with scene graphs

    Guangyao Zhai, Evin Pınar ¨Ornek, Shun-Cheng Wu, Yan Di, Federico Tombari, Nassir Navab, and Benjamin Busam. Commonscenes: Generating commonsense 3d indoor scenes with scene graphs. Advances in Neural Information Process- ing Systems (NeurIPS), 36, 2024. 3

    2024

  74. [82]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 3836–3847, 2023. 2, 3, 15, 17

    2023

  75. [83]

    Layoutdiffusion: Controllable diffu- sion model for layout-to-image generation

    Guangcong Zheng, Xianpan Zhou, Xuewei Li, Zhongang Qi, Ying Shan, and Xi Li. Layoutdiffusion: Controllable diffu- sion model for layout-to-image generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 22490–22499, 2023. 3

    2023

  76. [84]

    Bias in generative ai

    Mi Zhou, Vibhanshu Abhishek, Timothy Derdenger, Jaymo Kim, and Kannan Srinivasan. Bias in generative ai. arXiv preprint arXiv:2403.02726, 2024. 18

    2024 arXiv

  77. [85]

    The great wave

    Yi Zhou, Connelly Barnes, Jingwan Lu, Jimei Yang, and Hao Li. On the continuity of rotation representations in neu- ral networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5745–5753, 2019. 5, 13 12 LACONIC: A 3D Layout A...

    2019

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.