Pith. sign in

REVIEW 3 major objections 6 minor 56 references

Modeling each room as a simple Gaussian lets one diffusion model both generate and edit apartment floor plans under user constraints.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-10 13:19 UTC pith:BCGN2KCD

load-bearing objection Competent interactive floor-plan diffusion paper: GRE as a room-level state is a real, ablated idea, metrics on RPLAN are strong, and the isotropic limitation is real but already partly owned by the authors. the 3 major comments →

arxiv 2607.08086 v1 pith:BCGN2KCD submitted 2026-07-09 cs.CV

GRE-Diff: Gaussian Room Embeddings for Structured Layout Diffusion

Jing Wang , Haoran Xiong , Zihao Yan , Minglun Gong , Hui Huang This is my paper
classification cs.CV
keywords Floor PlanLayout GenerationControllable EditingDiffusion ModelsGaussian Room EmbeddingInteractive DesignPolygonal Layouts
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Apartment floor-plan design forces a designer to explore a huge space of room arrangements while keeping boundaries valid, rooms non-overlapping, and functional relationships intact. This paper claims that representing every room by an isotropic Gaussian—its center plus one scale number for extent—supplies a continuous room-level state that both initializes and steers a diffusion process. The resulting GRE-Diff system turns natural-language room counts or GUI sketches into diverse polygonal layouts, then supports local add/delete/move/anchor edits without regenerating everything from scratch. On the RPLAN residential dataset the method reports the best overall distributional match, highest coverage, and highest constraint satisfaction among the compared approaches. The practical goal is a single controllable pipeline that keeps AI suggestions and human refinement inside the same generative loop.

Core claim

The paper establishes that Gaussian Room Embeddings—each room cast as an isotropic two-dimensional Gaussian whose mean is the centroid and whose scalar variance captures spatial extent—form a continuous, order-robust geometric state that can initialize and parameterize a conditional diffusion process, thereby unifying high-quality polygonal floor-plan generation with structure-preserving interactive editing under boundary, room-type, and user-edit constraints.

What carries the argument

Gaussian Room Embedding (GRE): each room is a spatial Gaussian N(μ, σ²I); the predicted (μ, σ) set the initial noise and the room-dependent drift/variance of the diffusion, after which DenoisingNet recovers explicit polygons.

Load-bearing premise

A single center point and one size number per room are enough to steer diffusion into valid, functional apartment polygons even when boundaries are irregular or rooms are edited.

What would settle it

Run the model on a held-out collection of strongly non-rectilinear apartment outlines never seen in RPLAN; if FID, overlap-free rate or door-reachability collapse while room-count accuracy remains high, the isotropic GRE state is insufficient.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Users can obtain diverse, constraint-satisfying apartment polygons from natural-language room counts and sketched boundaries without hand-built graphs.
  • The same diffusion model supports add, delete, move, repurpose and anchor edits while freezing user-chosen rooms and preserving global coherence.
  • Reported metrics on RPLAN show lower FID/KID/MMD, higher coverage and higher boundary-plus-room F1 than the compared graph, GAN and diffusion baselines.
  • Automatic BC/RC feedback can resample failing layouts without external snapping or post-processing.
  • A dual-mode interface (LLM instructions or GUI operators) lets non-experts iterate designs inside one generative loop.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If GRE already unifies generation and editing, the same room-level state could serve as a shared interface for multi-turn agentic design loops that plan, check and re-edit layouts.
  • Failures on highly irregular contours suggest that anisotropic or multi-mode Gaussians may be required before the method generalizes beyond residential Manhattan-like plans.
  • Because GRE is compact and order-invariant, it is a natural candidate for coupling with downstream furniture-placement modules to close the empty-room-to-furnished-interior pipeline.
  • The same continuous room state may transfer to multi-story or commercial layouts once training data cover those typologies.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes GRE-Diff, a diffusion framework for controllable apartment floor-plan generation and interactive editing. Each room is represented by an isotropic Gaussian Room Embedding (GRE) with centroid μ and scalar scale σ; GuidanceNet predicts these embeddings from multi-modal conditions (room types/counts, boundary, optional polygons), and DenoisingNet denoises polygons initialized and steered by the GRE-parameterized reverse process. The system supports LLM-parsed natural-language constraints and five GUI editing operators (Add/Delete/Move/Repurpose/Anchor) with hard fixed-room masks. On RPLAN, GRE-Diff reports best or near-best FID/KID/MMD, highest COV, and highest F1 BC-RC among seven baselines, plus strong editing Tiny-ROE/F1, user preference over Graph2Plan and GT, and high overlap-free/reachability rates. Ablations contrast dual-path attention and GRE-as-state vs conditioning-only vs plain DDPM.

Significance. If the results hold under fair comparison, this is a useful systems contribution: a single diffusion model that unifies vectorized generation and localized editing under boundary and room-count constraints, with a practical dual-mode (LLM/GUI) interface. Strengths include multi-metric evaluation on a held-out RPLAN split, an explicit ablation showing that wiring GRE into the diffusion state transition (not merely as external conditioning) improves metrics (Table S1), functionality rates (overlap-free 95.56%, reachability 96.35%), a user preference study, and an honest limitations section with failure cases. The continuous room-level state idea is a reasonable alternative to ordered vertex sequences and discrete masks for structured layout diffusion.

major comments (3)
  1. Sec. 3.1–3.2 and Supp. C.2: The central representation claim is that an isotropic GRE N(μi, σi²I) is a sufficient room-level geometric state to initialize and parameterize polygon diffusion (forward/reverse Eqs. 1–3). Table S1 only contrasts full GRE-state vs conditioning-only vs plain DDPM; it does not test richer room states (anisotropic covariance, oriented box, multi-mode blob). Sec. 4.5 and Fig. 8 document failures precisely where isotropy is weakest (non-rectilinear/irregular boundaries, uneven vertex density, connectivity/semantics). Either add a controlled ablation with a richer room state under the same training budget, or substantially qualify the claim that GRE is a generally adequate geometric state rather than an effective prior for RPLAN’s largely rectilinear apartments.
  2. Table 1 and Sec. 4.2: Baseline fairness needs tighter documentation. HouseDiffusion is evaluated without boundary constraints by design; Graph2Plan’s strong numbers depend on retrieval-assisted post-processing (FID rises to 20.10 and RC falls to 50.19% without it). WallPLAN/GSDiff use snapping/alignment. Please report a primary comparison under a common protocol (same boundary+room-type conditions, no non-neural post-processing, or with post-processing clearly factored out for all methods) so that the headline ranking (FID 4.36, F1 BC-RC 99.21%) is not partly an artifact of unequal pipelines.
  3. Sec. 4.1–4.2 functionality metrics and Supp. on doors/windows: Overlap-free and reachability rates (95.56% / 96.35%) use annotated door labels to build connectivity graphs, while the method itself does not generate doors/windows and instead applies a post-hoc rule-based procedure (Supp. limitation 4). Clarify whether reachability is measured on model outputs alone or after rule-based opening insertion, and report connectivity without relying on GT door annotations if the claim is that generated layouts are functionally usable end-to-end.
minor comments (6)
  1. Fig. 7 caption refers to “Fig. S5” in the main text narrative (“Fig. S5 shows…”); renumber or fix the cross-reference so main-text figures are self-contained.
  2. Notation: the reverse update in Sec. 3.2 uses αt / ᾱt inconsistently with Supp. C.2; unify the noise schedule symbols and define ᾱt explicitly in the main text.
  3. Table 2 Tiny-ROE = 0.00% for all operations is striking; briefly state the LPIPS threshold and how “requested edit applied” is verified (semantic label check vs geometric IoU) so the metric is reproducible.
  4. User study (Sec. 4.2): report inter-rater agreement or confidence intervals on the 1740 responses; preference over GT is interesting and would benefit from a short discussion of possible confounds (rendering style, wall thickness).
  5. Related work: briefly position GRE against BlobGAN/BlobCtrl-style spatial Gaussians and PolyDiffuse set diffusion, which are cited but not contrasted on the room-state design choice.
  6. Typos/clarity: “load-bearing wall constraints” in Fig. 5 caption is not fully explained in the main text; define how load-bearing walls enter the conditioning y.

Circularity Check

0 steps flagged

No significant circularity: empirical generative method evaluated on held-out data with independent ablations.

full rationale

GRE-Diff is an empirical computer-vision paper that defines Gaussian Room Embeddings (isotropic N(μ, σ²I)), trains GuidanceNet and DenoisingNet on RPLAN, and reports standard external metrics (FID/KID/MMD/COV/BC/RC/F1, Tiny-ROE, LPIPS, user study) on held-out layouts. The reverse-process equations (Supp. C.2) parameterize diffusion with predicted GRE but do not algebraically force the reported similarity or controllability numbers; those are measured against real test polygons and baselines. The key ablation (Table S1) shows that wiring GRE into the diffusion state outperforms both plain DDPM and GRE-as-conditioning-only, which is independent empirical evidence rather than a tautology. No uniqueness theorem, fitted-parameter-as-prediction, or load-bearing self-citation chain reduces the central claims to their inputs by construction. Limitations on non-rectilinear boundaries are openly stated and do not create circularity.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central empirical claim rests on standard diffusion math, the RPLAN residential distribution, and the modeling choice that isotropic Gaussians plus transformer denoising suffice for structured layouts. Free parameters are ordinary ML hyperparameters and loss weights; the main invented entity is GRE itself, justified by ablations but without independent physical evidence outside this task.

free parameters (6)
  • Encoder loss weights λ1, λ2, λ3 = 1.0 / 1.0 / 0.5
    Set to 1.0, 1.0, 0.5 for Lperm, Lsem, Lreg; chosen for training stability, not derived.
  • Triplet soft-margin α = 0.1
    Margin in permutation-invariant triplet loss; hand-set.
  • GRE quantization bins K = 50
    Number of discrete bins for autoregressive (μx, μy, σ) prediction.
  • Learning rate and schedule = 2e-4
    AdamW 2e-4 with linear warm-up and cosine decay over 300 epochs; standard fitted training choice.
  • Polygon/boundary sampling resolution = 64 points
    Boundary encoded with 64 uniformly sampled points; affects geometric fidelity.
  • Editing coefficient γ = 0 or 1
    Switches polygonal prior contribution (0 generation, 1 editing); design hyperparameter.
axioms (5)
  • standard math Standard DDPM-style forward/reverse diffusion with noise schedules α_t, ᾱ_t is a valid generative process for 2D polygon vertices.
    Invoked throughout Sec. 3.2 and supplement C.2; reduces to ordinary DDPM when μ=0, σ=1.
  • domain assumption RPLAN vectorized residential plans are a representative distribution for evaluating apartment layout quality and constraints.
    All quantitative claims use RPLAN train/test splits (Sec. 4.1); generalization beyond this corpus is not established.
  • ad hoc to paper An isotropic Gaussian (μ, σI) adequately summarizes room location and extent for guiding polygon diffusion and editing.
    Core modeling choice in Sec. 3.1; full covariance or richer shape priors are not used; limitations section shows failures when geometry is irregular.
  • domain assumption LLM few-shot parsing of natural language into room counts/types and edit operators is reliable enough for the interactive pipeline.
    Sec. 3.3 and supplement B; no systematic error analysis of parse failures is reported.
  • domain assumption Doors and windows can be added by post-hoc geometric rules without joint learning.
    Explicitly stated in supplement limitations; reachability metric depends on annotated/derived doors.
invented entities (2)
  • Gaussian Room Embedding (GRE) no independent evidence
    purpose: Continuous room-level latent (centroid + isotropic scale) that initializes and parameterizes diffusion for generation and editing.
    Defined in Sec. 3.1; ablations (Table S1) support usefulness inside this model, but GRE has no independent measurement outside the paper’s training objective.
  • Dual-path GuidanceNet with generation/editing adapters no independent evidence
    purpose: Fuse semantic, boundary, and polygon tokens and switch between free generation and constrained editing.
    Architectural invention of the paper (Sec. 3.2); evaluated only within GRE-Diff.

pith-pipeline@v1.1.0-grok45 · 27756 in / 3642 out tokens · 39006 ms · 2026-07-10T13:19:39.586444+00:00 · methodology

0 comments
read the original abstract

Designing functional and aesthetically coherent floor plans requires exploring a vast space of possible room arrangements, a task that quickly becomes overwhelming for human designers. In this paper, we propose GRE-Diff, a controllable and interactive diffusion-based framework that automates the creation and editing of apartment floor plans under user-specified constraints. By combining AI-generated suggestions with real-time, human-in-the-loop editing, the system enables users to specify room types, room counts, boundary shapes, and editing operations through LLM-parsed instructions or GUI-based interaction. It then generates a diverse set of plausible and well-structured designs for refinement. At the core of our approach is Gaussian Room Embedding (GRE), a continuous latent representation that models each room as a spatial Gaussian distribution capturing its location and extent. Extensive experiments on the RPLAN dataset show that GRE-Diff produces high-quality, constraint-aware, and editable polygonal layouts, offering a practical step toward bridging AI-driven automation and human creativity in spatial design.

Figures

Figures reproduced from arXiv: 2607.08086 by Haoran Xiong, Hui Huang, Jing Wang, Minglun Gong, Zihao Yan.

Figure 1
Figure 1. Figure 1: Polygon layout and its Gaus￾sian Room Embedding. GRE provides a continuous room-level spatial state that initializes and guides polygon diffusion, while the final output remains an explicit polygonal layout. location, while \sigma _i summarizes its coarse geometric scale and spatial influence within the layout. GRE does not re￾place the explicit polygon representa￾tion. Instead, it serves as a compact room… view at source ↗
Figure 2
Figure 2. Figure 2: Inference process of GRE-Diff. GuidanceNet encodes semantic (Stoken), bound￾ary (Btoken), and polygon (Ptoken) conditions to predict Gaussian room embeddings (µi, σi). Samples drawn from N (µi, σ2 i I) serve as diffusion initialization and are it￾eratively refined by DenoisingNet under boundary constraints to generate vectorized layouts x i 0. The resulting layouts can be re-encoded as polygon tokens, enab… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of GRE-Diff, a controllable framework for floor plan generation and editing. Given apartment boundaries and room constraints, GuidanceNet predicts Gaussian room embeddings that guide the diffusion process, while DenoisingNet gen￾erates coherent polygonal layouts. The framework supports both language-based and GUI-based editing for interactive layout refinement. 3.3 User-guided Layout Generation an… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of floor plan generation. The red dashed regions high￾light failure cases in existing methods, often arising from inconsistencies in room num￾bers, functional mismatches, or inaccessible layouts. GRE-Diff generates structurally coherent and functionally consistent results, closely resembling GT layouts. geometries. In contrast, GRE-Diff embeds both boundary and room constraints direc… view at source ↗
Figure 5
Figure 5. Figure 5: Results of floor plans generated under various configurations. This figure shows the network’s adaptability in generating layouts with consistent room configurations and its flexibility in adjusting room arrangements within a fixed boundary, underscoring the impact of load-bearing wall constraints. The network autonomously optimizes room layouts while maintaining spatial organization and functionality. hib… view at source ↗
Figure 6
Figure 6. Figure 6: Novelty analysis using LPIPS. We plot the LPIPS distribution of 500 ran￾domly generated layouts and visualize rep￾resentative samples from different per￾centile groups together with their nearest retrieved training counterparts. [0.03, 0.05], indicating high percep￾tual alignment, while samples above 0.06 correspond to intentionally di￾verse geometric configurations. For il￾lustration, we display samples f… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison of floor plan editing. We apply editing operations to the original ground truth layout, as shown in the first row. Compared to other meth￾ods, our approach provides more precise control over room positions, ensuring that fixed structures (balcony or kitchen) remain unchanged while enabling flexible room relocation. Red dashed boxes mark incorrect edits from other methods. the best ba… view at source ↗
Figure 8
Figure 8. Figure 8: Representative limitation cases. Examples include (a) uneven vertex density; (b) incorrect room connectivity; (c) semantic mismatch; (d) limited generalization to non-rectilinear boundaries; (e-f) diverse layouts under identical non-Manhattan bound￾ary; (g) a tilted bedroom; and (h) correctly places two balconies along a sloped wall. Despite these, the layouts remain structurally coherent and spatially pla… view at source ↗
Figure 9
Figure 9. Figure 9: Snapshot of the intuitive and user-friendly floor plan design interface. 4.6 User Interface [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages · 9 internal anchors

  1. [1]

    In: International conference on machine learning

    Achlioptas, P., Diamanti, O., Mitliagkas, I., Guibas, L.: Learning representations and generative models for 3d point clouds. In: International conference on machine learning. pp. 40–49. PMLR (2018)

    work page 2018

  2. [2]

    In: Proc

    Bao,J.,Chen,D.,Wen,F.,Li,H.,Hua,G.:Cvae-gan:fine-grainedimagegeneration through asymmetric training. In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition. pp. 2745–2754 (2017)

    work page 2017

  3. [3]

    Demystifying MMD GANs

    Bińkowski, M., Sutherland, D.J., Arbel, M., Gretton, A.: Demystifying mmd gans. arXiv preprint arXiv:1801.01401 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  4. [4]

    In: Proc

    Brooks, T., Holynski, A., Efros, A.A.: Instructpix2pix: Learning to follow image editing instructions. In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition. pp. 18392–18402 (2023)

    work page 2023

  5. [5]

    Chen, J., Deng, R., Furukawa, Y.: Polydiffuse: Polygonal shape reconstruction via guided set diffusion models. Proc. Conf. on Neural Information Processing Systems 36(2024)

    work page 2024

  6. [6]

    In: Proc

    Chen, Q., Wu, Q., Tang, R., Wang, Y., Wang, S., Tan, M.: Intelligent home 3d: Automatic 3d-house design from linguistic descriptions only. In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition. pp. 12625–12634 (2020)

    work page 2020

  7. [7]

    Dhariwal, P., Nichol, A.: Diffusion models beat gans on image synthesis. Proc. Conf. on Neural Information Processing Systems34, 8780–8794 (2021)

    work page 2021

  8. [8]

    In: Proc

    Dupty,M.H.,Dong,Y.,Leng,S.,Fu,G.,Goh,Y.L.,Lu,W.,Lee,W.S.:Constrained layout generation with factor graphs. In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition. pp. 12851–12860 (2024)

    work page 2024

  9. [9]

    In: Proc

    Epstein, D., Park, T., Zhang, R., Shechtman, E., Efros, A.A.: Blobgan: Spatially disentangled scene representations. In: Proc. Euro. Conf. on Computer Vision. pp. 616–635. Springer (2022)

    work page 2022

  10. [10]

    In: Proc

    Gao, J., Shen, T., Wang, Z., Chen, W., Yin, K., Li, D., Litany, O., Gojcic, Z., Fidler, S.: Get3d: A generative model of high quality 3d textured shapes learned from images. In: Proc. Conf. on Neural Information Processing Systems (2022)

    work page 2022

  11. [11]

    In: Proc.IEEE/CVFConf.onComputerVision&PatternRecognition.pp.7793–7802 (2022)

    He, F., Huang, Y., Wang, H.: iplan: Interactive and procedural layout planning. In: Proc.IEEE/CVFConf.onComputerVision&PatternRecognition.pp.7793–7802 (2022)

    work page 2022

  12. [12]

    IEEE/CVF Conf

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition. pp. 770–778 (2016)

    work page 2016

  13. [13]

    Heusel,M.,Ramsauer,H.,Unterthiner,T.,Nessler,B.,Hochreiter,S.:Ganstrained by a two time-scale update rule converge to a local nash equilibrium. Proc. Conf. on Neural Information Processing Systems30(2017)

    work page 2017

  14. [14]

    In: Proceedings of the 32nd ACM International Con- ference on Multimedia

    Hong, S., Zhang, X., Du, T., Cheng, S., Wang, X., Yin, J.: Cons2plan: Vector floorplan generation from various conditions via a learning framework based on conditional diffusion models. In: Proceedings of the 32nd ACM International Con- ference on Multimedia. pp. 3248–3256 (2024)

    work page 2024

  15. [15]

    ACM Trans

    Hu, R., Huang, Z., Tang, Y., Van Kaick, O., Zhang, H., Huang, H.: Graph2plan: Learning floorplan generation from layout graphs. ACM Trans. on Graphics39(4), 118–1 (2020)

    work page 2020

  16. [16]

    In: Proc

    Hu, S., Wu, W., Wang, Y., Xu, B., Zheng, L.: Gsdiff: Synthesizing vector floor- plans via geometry-enhanced structural graph generation. In: Proc. AAAI Conf. on Artificial Intelligence. vol. 39, pp. 17323–17332 (2025) GRE-Diff 17

    work page 2025

  17. [17]

    Auto-Encoding Variational Bayes

    Kingma, D.P.: Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013)

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  18. [18]

    Tell2Design: A Dataset for Language-Guided Floor Plan Generation

    Leng, S., Zhou, Y., Dupty, M.H., Lee, W.S., Joyce, S.C., Lu, W.: Tell2design: A dataset for language-guided floor plan generation. arXiv preprint arXiv:2311.15941 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  19. [19]

    arXiv preprint arXiv:2503.13434 (2025)

    Li, Y., Li, L., Zhang, Z., Li, X., Wang, G., Li, H., Cun, X., Shan, Y., Zou, Y.: Blobctrl: A unified and flexible framework for element-level image generation and editing. arXiv preprint arXiv:2503.13434 (2025)

    work page arXiv 2025

  20. [20]

    In: Proc

    Liu, J., Xue, Y., Duarte, J., Shekhawat, K., Zhou, Z., Huang, X.: End-to-end graph-constrained vectorized floorplan generation with panoptic refinement. In: Proc. Euro. Conf. on Computer Vision. pp. 547–562. Springer (2022)

    work page 2022

  21. [21]

    Decoupled Weight Decay Regularization

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  22. [22]

    In: Proc

    Ma, X., Liu, Y., Gao, T., Huang, Q., Huang, H.: Clr-wire: Towards continuous latent representations for 3d curve wireframe generation. In: Proc. SIGGRAPH. pp. 77:1–77:11 (2025)

    work page 2025

  23. [23]

    In: Proc

    Ma, X., Liu, Y., Zhou, W., Wang, R., Huang, H.: Generating 3d house wireframes with semantics. In: Proc. Euro. Conf. on Computer Vision. vol. 15080, pp. 223–240 (2024)

    work page 2024

  24. [24]

    House-GAN++: Generative Adversarial Layout Refinement Networks

    Nauata, N., Hosseini, S., Chang, K., Chu, H., Cheng, C., Furukawa, Y.: House-gan++: generative adversarial layout refinement networks. arXiv preprint arXiv:2103.02574 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  25. [25]

    In: Proc

    Nauata, N., Chang, K.H., Cheng, C.Y., Mori, G., Furukawa, Y.: House-gan: Rela- tional generative adversarial networks for graph-constrained house layout genera- tion. In: Proc. Euro. Conf. on Computer Vision. pp. 162–177. Springer (2020)

    work page 2020

  26. [26]

    org/10.5281/zenodo.4957738,https://github.com/toshas/torch- fidelity, version: 0.3.0, DOI: 10.5281/zenodo.4957738

    Obukhov, A., Seitzer, M., Wu, P.W., Zhydenko, S., Kyl, J., Lin, E.Y.J.: High- fidelityperformancemetrics for generative models in pytorch (2020).https://doi. org/10.5281/zenodo.4957738,https://github.com/toshas/torch- fidelity, version: 0.3.0, DOI: 10.5281/zenodo.4957738

    work page doi:10.5281/zenodo.4957738 2020

  27. [27]

    In: Proc

    Para, W., Guerrero, P., Kelly, T., Guibas, L.J., Wonka, P.: Generative layout mod- eling using constraint graphs. In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition. pp. 6690–6700 (2021)

    work page 2021

  28. [28]

    Civil Engineering Design6(2), 41–52 (2024)

    Ploennigs, J., Berger, M.: Automating computational design with generative ai. Civil Engineering Design6(2), 41–52 (2024)

    work page 2024

  29. [29]

    ChatHouseDiffusion: Prompt-Guided Generation and Editing of Floor Plans

    Qin, S., He, C., Chen, Q., Yang, S., Liao, W., Gu, Y., Lu, X.: Chathouse- diffusion: Prompt-guided generation and editing of floor plans. arXiv preprint arXiv:2410.11908 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  30. [30]

    In: Proc

    Shabani, M.A., Hosseini, S., Furukawa, Y.: Housediffusion: Vector floorplan gen- eration via a diffusion model with discrete and continuous denoising. In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition. pp. 5466–5475 (2023)

    work page 2023

  31. [31]

    Springer (2015)

    Sully, A.: Interior Design: conceptual basis. Springer (2015)

    work page 2015

  32. [32]

    ACM Trans

    Sun, J., Wu, W., Liu, L., Min, W., Zhang, G., Zheng, L.: Wallplan: Synthesizing floorplans by learning to generate wall graphs. ACM Trans. on Graphics41(4), 1–14 (2022)

    work page 2022

  33. [33]

    Tang, H., Shao, L., Sebe, N., Van Gool, L.: Graph transformer gans with graph maskedmodelingforarchitecturallayoutgeneration.IEEETrans.PatternAnalysis & Machine Intelligence46(6), 4298–4313 (2024)

    work page 2024

  34. [34]

    Kimi K2: Open Agentic Intelligence

    Team, K., Bai, Y., Bao, Y., Chen, G., Chen, J., Chen, N., Chen, R., Chen, Y., Chen, Y., Chen, Y., et al.: Kimi k2: Open agentic intelligence. arXiv preprint arXiv:2507.20534 (2025) 18 J. Wang, H. Xiong, Z. Yan, M. Gong, and H. Huang

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  35. [35]

    In: 2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)

    Upadhyay, A., Dubey, A., Arora, V., Kuriakose, S.M., Agarawal, S.: Flnet: graph constrained floor layout generation. In: 2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW). pp. 1–6. IEEE (2022)

    work page 2022

  36. [36]

    Vaswani, A.: Attention is all you need. Proc. Conf. on Neural Information Process- ing Systems (2017)

    work page 2017

  37. [37]

    Word Representations via Gaussian Embedding

    Vilnis, L., McCallum, A.: Word representations via gaussian embedding. CoRR abs/1412.6623(2014),https://api.semanticscholar.org/CorpusID:13468104

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  38. [38]

    Automation in Construction154, 105036 (2023)

    Wang, L., Liu, J., Zeng, Y., Cheng, G., Hu, H., Hu, J., Huang, X.: Automated building layout generation using deep learning and graph algorithms. Automation in Construction154, 105036 (2023)

    work page 2023

  39. [39]

    Automation in Construction 140, 104385 (2022)

    Weber, R.E., Mueller, C., Reinhart, C.: Automated floorplan generation in archi- tectural design:A review ofmethods andapplications. Automation in Construction 140, 104385 (2022)

    work page 2022

  40. [40]

    ACM Trans

    Wu, W., Fu, X.M., Tang, R., Wang, Y., Qi, Y.H., Liu, L.: Data-driven interior plan generation for residential buildings. ACM Trans. on Graphics38(6) (2019)

    work page 2019

  41. [41]

    In: Proceedings of the 2025 International Conference on Multimedia Retrieval

    Xu, M., Lou, Y., Gao, X., Zhou, X.: Floorplan-diffusion: Automatic floor plan generation via pre-trained large latent diffusion model. In: Proceedings of the 2025 International Conference on Multimedia Retrieval. pp. 1617–1625 (2025)

    work page 2025

  42. [42]

    arXiv (2019)

    Yang, G., Huang, X., Hao, Z., Liu, M.Y., Belongie, S., Hariharan, B.: Pointflow: 3d point cloud generation with continuous normalizing flows. arXiv (2019)

    work page 2019

  43. [43]

    ACM Trans

    Yao, K., Zhang, L., Yan, X., Zeng, Y., Zhang, Q., Xu, L., Yang, W., Gu, J., Yu, J.: Cast: Component-aligned 3d scene reconstruction from an rgb image. ACM Trans. on Graphics44(4), 1–19 (2025)

    work page 2025

  44. [44]

    ImgEdit: A Unified Image Editing Dataset and Benchmark

    Ye, Y., He, X., Li, Z., Lin, B., Yuan, S., Yan, Z., Hou, B., Yuan, L.: Imgedit: A unified image editing dataset and benchmark. arXiv preprint arXiv:2505.20275 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  45. [45]

    In: Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

    Yin, J., Zeng, P., Sun, H., Dai, Y., Zheng, H., Zhang, M., Zhang, Y., Lu, S.: Floorplan-llama: Aligning architects’ feedback and domain knowledge in architec- tural floor plan generation. In: Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). pp. 6640– 6662 (2025)

    work page 2025

  46. [46]

    In: Proc

    Zhang, H., Savov, A., Dillenburger, B.: Maskplan: Masked generative layout plan- ning from partial input. In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition. pp. 8964–8973. IEEE (2024)

    work page 2024

  47. [47]

    Zhang, K., Mo, L., Chen, W., Sun, H., Su, Y.: Magicbrush: A manually annotated dataset for instruction-guided image editing. Proc. Conf. on Neural Information Processing Systems36, 31428–31449 (2023)

    work page 2023

  48. [48]

    In: Proc

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: Proc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition (2018)

    work page 2018

  49. [49]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Zhou, L., Du, Y., Wu, J.: 3d shape generation and completion through point-voxel diffusion. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 5826–5835 (October 2021)

    work page 2021

  50. [50]

    three bedroom, one living room, one kitchen, one bathroom

    Zong, Z., Zhan, Z., Tan, G.: Housellm: Llm-assisted two-phase text-to-floorplan generation. arXiv e-prints pp. arXiv–2411 (2024) GRE-Diff 19 Supplementary Material A INTRODUCTION This supplementary document provides detailed information on various aspects of theGRE-Diffpaper, including the LLM prompt in Section B, framework details in Section C, evaluatio...

    work page 2024

  51. [51]

    Move room <index> to position (x,y)

    work page

  52. [52]

    Add a new <room_type> room at position (x,y)

    work page

  53. [53]

    Think: The user wants to move a bathroom

    Change room <index> to <room_type> Examples Condition: [{index: 1, type: Living Room, area: 5394, center: [103, 120]}, {index: 2, type: Bedroom, area: 2684, center: [176, 89]}, {index: 3, type: Kitchen, area: 1178, center: [66, 86]}, {index: 4, type: Bathroom, area: 762, center: [125, 86]}, {index: 5, type: Bathroom, GRE-Diff 21 area: 400, center: [95, 80...

    work page 1911

  54. [54]

    A room embedding should be invariant to polygonal vertex permutations and remain consistent under forward-diffused variants

    Permutation LossL perm. A room embedding should be invariant to polygonal vertex permutations and remain consistent under forward-diffused variants. To achieve this, we adopt a bidirectional triplet loss that pulls em- beddings of the same room closer while pushing embeddings of other rooms apart. For roomi, the triplet objectives are: LTri.(xi t, X0) = m...

    work page

  55. [55]

    Room embeddings should encode not only geom- etry but also semantics

    Semantic LossLsem. Room embeddings should encode not only geom- etry but also semantics. To this end, we predict semantic labels directly from encoder features and supervise them using a Cross-Entropy loss: Lsem = NX i=1 (− CX c=1 Sgt(i, c)·log eSpred(i,c) PC j=1 eSpred(i,j) ), whereNdenotes the number of rooms,Cis the total number of semantic cat- egorie...

    work page

  56. [56]

    Add a balcony

    Regularization LossL reg. To maintain numerical stability and avoid pathological variances, we regularize the predicted Gaussian parameters as: Lreg = NX i=1 (||µi||2 +|| 1 σi ||2). Theregularizerconstrainstheembeddingstoremainnumericallywell-behaved, avoiding excessive mean offsets and preventing extremely small or large variances that may impair diffusi...

    work page