Pith. sign in

REVIEW 3 major objections 6 minor 44 references

Top2Pano: Learning to Generate Indoor Panoramas from Top-Down View

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 Top2Pano claims to be the first framework that turns 2D top-down views of rooms into photorealistic 360° indoor panoramas, beating existing baselines on Matterport3D and Gibson.

desk verdict New task with a plausible but under-specified pipeline; the unstated height/scale dependency undercuts the generalization claim. read the letter →

arxiv 2507.21371 v1 pith:TKS55ECG submitted 2025-07-28 cs.CV

Zitong Zhang , Suranjan Gautam , Rui Yu This is my paper · ORCID
classification cs.CV
keywords panoramagenerationtop-downviewvolumetricoccupancydiffusionmodelsControlNetindoorscenesynthesisfloorplangeneralizationcross-view
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 proposes Top2Pano, a three-stage pipeline that takes a 2D top-down view of a room—like a floorplan—and generates a photorealistic 360° panorama of what that room would look like from inside. Its central claim is that no prior work has done this directly from top-down views, and that the method outperforms existing baselines on two indoor datasets in both image quality metrics and structural consistency. A sympathetic reader should care because the input is lightweight and semantic-free: no object labels, no 3D mesh, just an overhead image plus a camera position. If the claim holds, applications like virtual walkthroughs from architectural blueprints, interior-design style exploration, and robot spatial reasoning become practical.

What carries the argument

The load-bearing object is an estimated volumetric occupancy map $V_{\mathrm{occ}}$, a stack of vertical voxel layers learned from the top-down image and its SAM segmentation by the OccRecon module. This occupancy map is rendered with volumetric ray marching ($\alpha$-composited depth and color along each ray) into coarse equirectangular depth and color panoramas, and those two coarse panoramas become the conditioning inputs to PanoGen, a ControlNet-based diffusion module that produces the final image. The occupancy map carries the geometry, the coarse render carries the top-down correspondence, and the diffusion stage supplies the missing appearance and fine texture. Structural reinforcement on walls and floor, plus an alignment loss and a color histogram loss, keep the final output consistent with the inferred layout.

What would settle it

Take a fixed top-down image of a real room, generate a panorama, then feed the same image scaled by 50% or with a different stated ceiling height while holding the camera position fixed; if the output panorama's depth and wall positions do not change in a way consistent with the altered scale or height, the coordinate-mapping assumption is falsified.

Watch

Extended reading notes

Core claim

Top2Pano's central discovery is that a two-stage coarse-to-fine scheme resolves the ambiguous task of inferring 3D structure from a flat top-down image: first learn a volumetric occupancy map from the top-down view and its segmentation, render coarse depth and color panoramas from that occupancy via volumetric ray marching, then let a diffusion-based refinement module add photorealism while preserving the coarse geometry. The paper reports that the full model achieves PSNR 11.72 / SSIM 0.4409 / FID 30.84 / LPIPS 0.6029 on Matterport3D and PSNR 11.58 / SSIM 0.4851 / FID 28.68 / LPIPS 0.6282 on Gibson, beating the best baseline in every metric, with the largest margins in FID. It also reports that the method generalizes to schematic and hand-drawn floorplans and supports text- or image-guided style transfer with less than five style images.

Load-bearing premise

The method assumes the room's physical height and metric scale are known in advance, since it converts top-down pixel coordinates into 3D occupancy coordinates using those quantities, but it never explains how an arbitrary floorplan supplies them.

Editorial extensions

If this is right

  • A floorplan or even a hand-drawn sketch becomes a sufficient input for photorealistic first-person walkthroughs, since the method generalizes to schematic plans without semantic labels.
  • Cross-dataset results (Gibson to Matterport3D and vice versa) keep most metric performance, suggesting the learned occupancy and refinement stages transfer across scene distributions rather than memorizing one dataset.
  • Because the coarse depth condition is always enforced, users can change style with text or a few example images without collapsing the room's geometry; stylization and structure are decoupled up to a tuneable tradeoff.
  • Adding a rectangle to a floorplan changes the generated panorama (e.g., producing a washstand with a mirror), so the pipeline supports object-level scene editing by editing the top-down input.

Reading between the lines

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

  • The paper's large FID improvements relative to its small PSNR/SSIM improvements suggest the real gain is perceptual and texture realism rather than pixel-accurate reconstruction; readers should weight FID and LPIPS accordingly when judging the method.
  • The unstated height and scale dependency implies that a scale- or height-prediction front end, or canonicalization of arbitrary floorplans, will be needed before the method can be used on unconstrained user-drawn plans.
  • The same coarse-to-fine recipe—learned occupancy, volumetric render, then diffusion refinement—should transfer to other sparse-to-panorama settings such as satellite-to-street-view or outdoor plan-to-view synthesis, where the vertical dimension is equally ambiguous.
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. The paper introduces Top2Pano, a system that generates 360° indoor panoramas from a single 2D top-down view. The pipeline first applies SAM to segment the top-down view, then a diffusion-based OccRecon module estimates a volumetric occupancy map; volumetric rendering converts the occupancy map into coarse depth and color equirectangular panoramas; finally, a ControlNet-based PanoGen module refines these coarse images into photorealistic panoramas, optionally with text- or image-guided stylization. Experiments on Matterport3D and Gibson report quantitative gains over Sat2Density-based and PanFusion baselines, ablations of the main components, cross-dataset results, and qualitative demonstrations on schematic and hand-drawn floorplans. The central claims are that Top2Pano outperforms existing methods in image quality and structural consistency and generalizes to floorplan-like inputs.

Significance. The task is timely and the pipeline is a sensible coarse-to-fine solution. The explicit use of learned occupancy plus volumetric rendering gives the method a geometric grounding that pure text-to-image or image-to-image baselines lack, and the large FID improvements (about 55 and 51 points on the two datasets) are promising. The paper is also candid about failure cases and provides qualitative evidence for stylization and floorplan generalization. At the same time, the quantitative evidence is presented without variance or significance testing, and the key input dependency on known room height and scale is not addressed. The contribution is potentially useful, but the strength of the current claims exceeds what the experiments demonstrate; the result should be accepted only after the input dependency and evaluation protocol are fixed.

major comments (3)
  1. [§3.1–3.2, Eqs. (2)–(3)] The coarse-rendering stage assumes that the physical scale and room height of the input are known, but the paper never explains where these quantities come from for a general input. Section 3.1 states 'Given the scale information of the scene' and Section 3.2 states 'With the room height known,' yet no estimation procedure or user-supplied parameter is described. For dataset top-down views rendered from 3D meshes these values are available, but for the schematic and hand-drawn floorplans used to support the generalization claim (§4.6, Figures 1 and 5) they are not. If height or scale is incorrect, the occupancy queries and the compositing in Eqs. (2) and (3) occur in a mis-scaled volume, so wall positions, furniture heights, floor boundaries, and occlusions are systematically wrong. This is a load-bearing input dependency: please specify how height and scale are obtained (or estimated), report sensitivity to their values, and qualify the generalization claim accordingly. The limitation discussion in §4.7 should also mention this dependency.
  2. [§4.4, Table 2] The quantitative comparison is reported as single-run point estimates with no error bars, confidence intervals, or significance tests. The PSNR advantage over the strongest baseline is 0.27 dB on Matterport3D and 0.22 dB on Gibson, well within typical run-to-run variation for stochastic diffusion sampling, while the FID differences are large. The claim that Top2Pano 'outperforms baselines' is not supportable from a single seed. Please report mean±std over at least three seeds (ideally with a significance test) for Tables 2 and 4, and state the number of samples used for FID.
  3. [§4.4] The description of baseline adaptation is insufficient for an apples-to-apples comparison. Sat2Density is a satellite-to-ground method and PanFusion is a text-to-panorama method, but the paper does not state what inputs these baselines receive (e.g., the same top-down view, camera position, height/scale, text prompt, or segmentation) or how their hyperparameters were selected. Please specify the exact input and conditioning setup for each baseline, and consider including a layout-guided method such as ControlRoom3D or Ctrl-Room (cited as closely related in §2.2) to position the method against the most relevant prior work.
minor comments (6)
  1. [Table 3] The ablation table is difficult to read because the checkmark columns are not aligned with the header labels and the rows are not named; please add row labels such as 'w/o floor', 'w/o wall', 'w/o segment', 'w/o depth', and 'w/o color' as used in Figure A.1.
  2. [§4.6, Figure 5] The text says the hand-drawn floorplan sketch is shown in 'Figure 5, first row', but the caption assigns hand-drawn sketches to the second row; correct the reference.
  3. [§3.2] Please define the pixel-to-physical ratio and the fixed ray length; currently 'we first compute the ratio' and 'fixed-length ray' are mentioned without equations or units, which prevents reproduction.
  4. [§3.3, Eq. (4)] Equation (4) writes the PanoGen output as a function of only Icolor and Idepth, but Section 3.3 and Figure 2 also condition on text; clarify the conditioning notation.
  5. [§3.5, Eq. (11)] In the LoRA description, the dimension d is not defined as the dimension of the adapted weight matrices; state which weight matrices are adapted and how α in Eq. (12) is set.
  6. [§4.3] The reported '3.3 billion parameters' likely refers to the full diffusion backbone plus ControlNet clone, but it would help to state how many parameters are trainable versus frozen.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Top2Pano's occupancy-to-rendering-to-refinement chain is trained and evaluated against external ground-truth panoramas, with no fitted parameter renamed as a prediction.

full rationale

No circular step is present. The paper's derivation chain is: input top-down image and SAM segmentation are fed to OccRecon to estimate volumetric occupancy (Eq. 1); volumetric rendering produces coarse depth (Eq. 2) and coarse color (Eq. 3); PanoGen, a ControlNet-based diffusion module, refines these coarse representations into the final panorama (Eq. 4). The final output is supervised by external ground-truth panoramas through the denoising MSE loss, alignment loss, and histogram color loss (Eqs. 5, 6, 9, and 10), and is compared against baselines using external metrics (PSNR, SSIM, FID, LPIPS). Although the coarse color panorama is constructed by directly projecting colors from the top-down view, this is an intermediate conditioning signal, not a claimed prediction; the final panorama is not equivalent to that projection and is scored against ground-truth images. No parameter is fitted to a subset of the target data and then reported as a prediction. No uniqueness theorem or load-bearing claim is imported from the authors' prior work; the self-citations in the related-work section are contextual and do not justify the central result. The reviewer's concern about room height and scale in Sections 3.1 and 3.2 is a real input-dependency and generalization limitation, but it is not circularity: the paper does not define the target panoramas, the evaluation metrics, or the baselines in terms of the height or scale parameters. The pipeline is self-contained as a learned cross-view synthesis method trained and tested against external ground-truth data, so the appropriate circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim does not introduce new physical entities or fitted constants. It rests mainly on input assumptions about known room scale and height, the reliability of SAM segmentation on top-down views, and the sufficiency of top-down color for downstream diffusion. The only hand-chosen numeric factors (ray length scale, LoRA rank, histogram bins) are rendering and training details.

free parameters (3)
  • coarse color ray length scale = 0.5
    Section 3.2 halves the ray length for coarse color sampling after observing banding artifacts; this hand-chosen factor affects rendering quality of the floor texture but is not a fitted scientific constant.
  • LoRA rank = 8
    Section 3.5 sets r=8 for style adaptation, adding 0.8% new parameters; this is optional stylization and not central to the main panorama generation claim.
  • histogram bins = 256
    Equation (7) uses 256 bins for the color histogram loss; a standard training detail.
assumptions (5)
  • ad hoc to paper Room height and physical scale are known for each input top-down view.
    Sections 3.1 and 3.2 state 'Given the scale information of the scene' and 'With the room height known' but never describe how these are obtained, especially for schematic or hand-drawn floorplans.
  • domain assumption SAM segmentation provides reliable room-boundary and furniture cues on indoor top-down views.
    Section 3.1 reports that current semantic segmentation models struggle on top-down views, so SAM is used, but no quantitative evidence of SAM segmentation quality on top-down inputs is provided.
  • domain assumption Wall voxels can be reliably identified and forced to occupancy 1.
    Section 3.1 solidifies wall voxels by applying the maximum value, which presumes walls are visible in the top-down boundaries and that forced occupancy improves geometry.
  • domain assumption Coarse colors sampled from the top-down view are sufficient to guide vertical-surface appearance through diffusion.
    Section 3.2 projects colors directly from the top-down image; walls, windows, and furniture faces are not directly visible, so the diffusion stage must hallucinate them plausibly.
  • domain assumption Orthographic top-down renderings of textured 3D meshes are representative of real floorplans and hand-drawn sketches.
    Section 3.5 and Section 4.1 justify training on orthographic views to improve floorplan generalization, but no quantitative distribution-shift analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Top2Pano: Learning to Generate Indoor Panoramas from Top-Down View." pith.science (2026). https://pith.science/paper/TKS55ECG

@misc{pith2026250721371,
  author       = {Pith},
  title        = {Pith review of: Top2Pano: Learning to Generate Indoor Panoramas from Top-Down View},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TKS55ECG}},
  note         = {Machine review of arXiv:2507.21371}
}
read the original abstract

Generating immersive 360{\deg} indoor panoramas from 2D top-down views has applications in virtual reality, interior design, real estate, and robotics. This task is challenging due to the lack of explicit 3D structure and the need for geometric consistency and photorealism. We propose Top2Pano, an end-to-end model for synthesizing realistic indoor panoramas from top-down views. Our method estimates volumetric occupancy to infer 3D structures, then uses volumetric rendering to generate coarse color and depth panoramas. These guide a diffusion-based refinement stage using ControlNet, enhancing realism and structural fidelity. Evaluations on two datasets show Top2Pano outperforms baselines, effectively reconstructing geometry, occlusions, and spatial arrangements. It also generalizes well, producing high-quality panoramas from schematic floorplans. Our results highlight Top2Pano's potential in bridging top-down views with immersive indoor synthesis.

Figures

Figures reproduced from arXiv: 2507.21371 by the authors.

Figure 1
Figure 1. Top: We present Top2Pano, a method for synthesizing high-quality indoor panoramas from a top-down view. Given a camera position, Top2Pano generates panoramas that are both visually compelling and geometrically accurate. Bottom: Our model demonstrates strong generalization capabilities. When provided with schematic floor plans as input, Top2Pano produces photorealistic and structurally coherent panoramas. Additionall… view at source ↗
Figure 2
Figure 2. Overview of the proposed Top2Pano pipeline. The pipeline begins by segmenting the top-down view using SAM [19]. Both the segmented top-down image and the original top-down view are then processed by the OccRecon module to estimate the scene’s 3D volumetric occupancy. Next, given the camera position, the system employs volumetric rendering to generate coarse depth and color panoramas. These coarse images are subseque… view at source ↗
Figure 3
Figure 3. Qualitative comparisons on the Matterport3D dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Qualitative comparisons on the Gibson dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 6
Figure 6. Figure 6: Top2Pano enables panorama manipulation via composi [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Failure cases (zoom in to view error types). [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 40 canonical work pages

  1. [1]

    Di- verse plausible 360-degree image outpainting for efficient 3dcg background creation

    Naofumi Akimoto, Yuhi Matsuo, and Yoshimitsu Aoki. Di- verse plausible 360-degree image outpainting for efficient 3dcg background creation. In IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, CVPR, pages 11431– 11440, 2022. 3

    work page 2022

  2. [2]

    Guibas, and Andrea Tagliasacchi

    Sherwin Bahmani, Jeong Joon Park, Despoina Paschalidou, Xingguang Yan, Gordon Wetzstein, Leonidas J. Guibas, and Andrea Tagliasacchi. CC3D: layout-conditioned generation of compositional 3d scenes. In IEEE/CVF International Con- ference on Computer Vision, ICCV, pages 7137–7147, 2023. 1, 3

    work page 2023

  3. [3]

    Matterport3d: Learning from rgb-d data in indoor environments

    Angel Chang, Angela Dai, Thomas Funkhouser, Maciej Hal- ber, Matthias Niessner, Manolis Savva, Shuran Song, Andy Zeng, and Yinda Zhang. Matterport3d: Learning from rgb-d data in indoor environments. International Conference on 3D Vision (3DV), 2017. 1, 6

    work page 2017

  4. [4]

    Layout2scene: 3d semantic layout guided scene generation via geometry and appearance diffusion pri- ors

    Minglin Chen, Longguang Wang, Sheng Ao, Ye Zhang, Kai Xu, and Yulan Guo. Layout2scene: 3d semantic layout guided scene generation via geometry and appearance diffusion pri- ors. arXiv preprint arXiv:2501.02519, 2025. 1, 3

    arXiv 2025

  5. [5]

    Text2light: Zero-shot text-driven hdr panorama generation

    Zhaoxi Chen, Guangcong Wang, and Ziwei Liu. Text2light: Zero-shot text-driven hdr panorama generation. ACM Trans- actions on Graphics (TOG), 41(6):1–16, 2022. 3

    work page 2022

  6. [6]

    Guided co-modulated gan for 360 ° field of view extrapolation

    Mohammad Reza Karimi Dastjerdi, Yannick Hold-Geoffroy, Jonathan Eisenmann, Siavash Khodadadeh, and Jean-Franc ¸ois Lalonde. Guided co-modulated gan for 360 ° field of view extrapolation. In 2022 International Conference on 3D Vision (3DV), pages 475–485, 2022. 3

    work page 2022

  7. [7]

    A density-based algorithm for discovering clusters in large spatial databases with noise

    Martin Ester, Hans-Peter Kriegel, J¨org Sander, and Xiaowei Xu. A density-based algorithm for discovering clusters in large spatial databases with noise. In Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD), pages 226–231, 1996. 6

    work page 1996

  8. [8]

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

    Chuan Fang, Xiaotao Hu, Kunming Luo, and Ping Tan. Ctrl- room: Controllable text-to-3d room meshes generation with layout constraints. arXiv preprint arXiv:2310.03602, 2023. 1, 3

    arXiv 2023

Show all 44 references
  1. [9]

    Prim2room: Layout-controllable room mesh generation from primitives

    Chengzeng Feng, Jiacheng Wei, Cheng Chen, Yang Li, Pan Ji, Fayao Liu, Hongdong Li, and Guosheng Lin. Prim2room: Layout-controllable room mesh generation from primitives. arXiv preprint arXiv:2409.05380, 2024. 1, 3

    2024 arXiv

  2. [10]

    Diffusion360: Seamless 360 degree panoramic im- age generation based on diffusion models

    Mengyang Feng, Jinlin Liu, Miaomiao Cui, and Xuansong Xie. Diffusion360: Seamless 360 degree panoramic im- age generation based on diffusion models. arXiv preprint arXiv:2311.13141, 2023. 1, 3

    2023 arXiv

  3. [11]

    What’s in my room? object recognition on indoor panoramic images

    Julia Guerrero-Viu, Clara Fernandez-Labrador, C ´edric De- monceaux, and Jos ´e Jes ´us Guerrero. What’s in my room? object recognition on indoor panoramic images. In IEEE International Conference on Robotics and Automation, ICRA, pages 567–573, 2020. 1

    2020

  4. [12]

    Where are you? localization from embodied dialog

    Meera Hahn, Jacob Krantz, Dhruv Batra, Devi Parikh, James M Rehg, Stefan Lee, and Peter Anderson. Where are you? localization from embodied dialog. arXiv preprint arXiv:2011.08277, 2020. 6

    2011 arXiv

  5. [13]

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

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bern- hard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In Annual Conference on Neural Information Processing Sys- tems, pages 6626–6637, 2017. 6

    2017

  6. [14]

    LoRA: Low-rank adaptation of large language models

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

  7. [15]

    Scene4u: Hierarchical layered 3d scene reconstruction from single panoramic image for your immerse exploration

    Zilong Huang, Jun He, Junyan Ye, Lihan Jiang, Weijia Li, Yiping Chen, and Ting Han. Scene4u: Hierarchical layered 3d scene reconstruction from single panoramic image for your immerse exploration. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 26...

  8. [16]

    Duospacenet: Leveraging both bird’s-eye-view and perspective view representations for 3d object detection

    Zhe Huang, Yizhe Zhao, Hao Xiao, Chenyan Wu, and Lingt- ing Ge. Duospacenet: Leveraging both bird’s-eye-view and perspective view representations for 3d object detection. In IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPR Workshops, pages 2560–2570,

  9. [17]

    Cubediff: Repurposing diffusion-based image models for panorama generation

    Nikolai Kalischek, Michael Oechsle, Fabian Manhardt, Philipp Henzler, Konrad Schindler, and Federico Tombari. Cubediff: Repurposing diffusion-based image models for panorama generation. In International Conference on Learn- ing Representations, ICLR, 2025. 3

    2025

  10. [18]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In International Conference on Learning Representations, ICLR, 2015. 6

    2015

  11. [19]

    Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross B

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chlo´e Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C. Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross B. Girshick. Segment anything. In IEEE/CVF Interna- tional Conference on Computer Vision, ICCV, pa...

    2023

  12. [20]

    Computer-aided layout generation for building design: A review

    Jiachen Liu, Yuan Xue, Haomiao Ni, Rui Yu, Zihan Zhou, and Sharon X Huang. Computer-aided layout generation for building design: A review. arXiv preprint arXiv:2504.09694,

    arXiv

  13. [21]

    Huang, and Hengkai Guo

    Jiachen Liu, Rui Yu, Sili Chen, Sharon X. Huang, and Hengkai Guo. Towards in-the-wild 3d plane reconstruction from a single image. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 27027–27037,

  14. [22]

    Oswald, Marc Pollefeys, and Rongjun Qin

    Xiaohu Lu, Zuoyue Li, Zhaopeng Cui, Martin R. Oswald, Marc Pollefeys, and Rongjun Qin. Geometry-aware satellite- to-ground image synthesis for urban areas. InIEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), pages 856–864, 2020. 3

    2020

  15. [23]

    Refocusable gigapixel panoramas for immersive VR experiences

    Wentao Lyu, Peng Ding, Yingliang Zhang, Anpei Chen, Minye Wu, Shu Yin, and Jingyi Yu. Refocusable gigapixel panoramas for immersive VR experiences. IEEE Trans. Vis. Comput. Graph., 27(3):2028–2040, 2021. 1

    2021

  16. [24]

    Nelson L. Max. Optical models for direct volume rendering. IEEE Trans. Vis. Comput. Graph., 1(2):99–108, 1995. 4

    1995

  17. [25]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- 9 thesis. In European Conference on Computer Vision, ECCV, pages 405–421, 2020. 4

    2020

  18. [26]

    Atiss: Autoregres- sive transformers for indoor scene synthesis

    Despoina Paschalidou, Amlan Kar, Maria Shugrina, Karsten Kreis, Andreas Geiger, and Sanja Fidler. Atiss: Autoregres- sive transformers for indoor scene synthesis. In Advances in Neural Information Processing Systems (NeurIPS), 2021. 3

    2021

  19. [27]

    Deep scene synthesis of atlanta-world interiors from a single omnidirectional image

    Giovanni Pintore, Fabio Bettio, Marco Agus, and Enrico Gobbetti. Deep scene synthesis of atlanta-world interiors from a single omnidirectional image. IEEE Trans. Vis. Comput. Graph., 29(11):4708–4718, 2023. 1

    2023

  20. [28]

    Sat2density: Faithful density learning from satellite-ground image pairs

    Ming Qian, Jincheng Xiong, Gui-Song Xia, and Nan Xue. Sat2density: Faithful density learning from satellite-ground image pairs. In IEEE/CVF International Conference on Com- puter Vision, ICCV, pages 3660–3669, 2023. 3, 4, 6, 7

    2023

  21. [29]

    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 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, CVPR, pages 10674–10685, 2022. 6

    2022

  22. [30]

    Controlroom3d: Room generation 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, Peizhao Zhang, Bastian Leibe, Peter Vajda, and Ji Hou. Controlroom3d: Room generation using semantic proxy rooms. In IEEE Conference on Computer...

    2024

  23. [31]

    Geometry-guided street-view panorama synthesis from satel- lite imagery

    Yujiao Shi, Dylan Campbell, Xin Yu, and Hongdong Li. Geometry-guided street-view panorama synthesis from satel- lite imagery. IEEE Trans. Pattern Anal. Mach. Intell., 44(12): 10009–10022, 2022. 3

    2022

  24. [32]

    Planarsplat- ting: Accurate planar surface reconstruction in 3 minutes

    Bin Tan, Rui Yu, Yujun Shen, and Nan Xue. Planarsplat- ting: Accurate planar surface reconstruction in 3 minutes. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 1190–1199, 2025. 1

    2025

  25. [33]

    Chang, and Manolis Savva

    Madhawa Vidanapathirana, Qirui Wu, Yasutaka Furukawa, Angel X. Chang, and Manolis Savva. Plan2scene: Converting floorplans to 3d scenes. In IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, CVPR, pages 10733– 10742, 2021. 3

    2021

  26. [34]

    360-degree panorama generation from few unregistered nfov images

    Jionghao Wang, Ziyu Chen, Jun Ling, Rong Xie, and Li Song. 360-degree panorama generation from few unregistered nfov images. In Proceedings of the 31st ACM International Conference on Multimedia, MM, pages 6811–6821, 2023. 3

    2023

  27. [35]

    Cross-view panorama image synthesis

    Songsong Wu, Hao Tang, Xiao-Yuan Jing, Haifeng Zhao, Jianjun Qian, Nicu Sebe, and Yan Yan. Cross-view panorama image synthesis. IEEE Trans. Multim., 25:3546–3559, 2023. 3

    2023

  28. [36]

    Zamir, Zhi-Yang He, Alexander Sax, Ji- tendra Malik, and Silvio Savarese

    Fei Xia, Amir R. Zamir, Zhi-Yang He, Alexander Sax, Ji- tendra Malik, and Silvio Savarese. Gibson env: real-world perception for embodied agents. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR, 2018. 6

    2018

  29. [37]

    Geospecific view generation geometry-context aware high-resolution ground view infer- ence from satellite views

    Ningli Xu and Rongjun Qin. Geospecific view generation geometry-context aware high-resolution ground view infer- ence from satellite views. In European Conference on Com- puter Vision, ECCV, pages 349–366, 2024. 3

    2024

  30. [38]

    Scenecraft: Layout-guided 3d scene generation

    Xiuyu Yang, Yunze Man, Jun-Kun Chen, and Yu-Xiong Wang. Scenecraft: Layout-guided 3d scene generation. In Advances in Neural Information Processing Systems, 2024. 1, 3

    2024

  31. [39]

    Diffpano: Scalable and consistent text to panorama generation with spherical epipolar-aware diffusion

    Weicai Ye, Chenhao Ji, Zheng Chen, Junyao Gao, Xiaoshui Huang, Song-Hai Zhang, Wanli Ouyang, Tong He, Cairong Zhao, and Guofeng Zhang. Diffpano: Scalable and consistent text to panorama generation with spherical epipolar-aware diffusion. In Annual Conference on Neural Informat...

    2024

  32. [40]

    Rui Yu, Jiachen Liu, Zihan Zhou, and Sharon X. Huang. Nerf-enhanced outpainting for faithful field-of-view extrap- olation. In IEEE International Conference on Robotics and Automation, ICRA, pages 16826–16833, 2024. 1

    2024

  33. [41]

    Taming stable diffusion for text to 360° panorama image gen- eration

    Cheng Zhang, Qianyi Wu, Camilo Cruz Gambardella, Xi- aoshui Huang, Dinh Phung, Wanli Ouyang, and Jianfei Cai. Taming stable diffusion for text to 360° panorama image gen- eration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024. 1, 3, 6, 7

    2024

  34. [42]

    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 IEEE/CVF International Conference on Computer Vision, ICCV, pages 3813–3824, 2023. 2, 3, 4, 6

    2023

  35. [43]

    Efros, Eli Shechtman, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In IEEE Conference on Com- puter Vision and Pattern Recognition, CVPR, pages 586–595,

  36. [44]

    Panorama genera- tion from nfov image done right

    Dian Zheng, Cheng Zhang, Xiao-Ming Wu, Cao Li, Chengfei Lv, Jian-Fang Hu, and Wei-Shi Zheng. Panorama genera- tion from nfov image done right. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 21610–21619, 2025. 1 10 A. Qualitative Results on Ablat...

    2025

Pith tools

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