Pith. sign in

REVIEW 4 major objections 6 minor 52 references

Incomplete design sketches plus a fixed 2D floor-plan contract beat fully specified baselines at both 2D layout and furnished 3D residential scenes.

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

2026-07-30 20:55 UTC pith:LLKMGKVV

load-bearing objection Solid progressive floor-plan + grounded furnishing system; the 61% FID win is real, but the “25% sketch beats all baselines” line is an unfair comparison and should not be the headline. the 4 major comments →

arxiv 2607.23491 v1 pith:LLKMGKVV submitted 2026-07-26 cs.CV cs.CL

PlanCraft: Sketch, Refine, and Furnish for Architect-Inspired Progressive 3D Residential Scene Generation

Pengyu Zeng , Yuqin Dai , Jun Yin , Ziyang Han , Ng Cheuk Hei , Jing Zhong , Chaoyang Shi , ZhanXiang Jin
show 3 more authors
Maowei Jiang Yuxing Han Shuai Lu
This is my paper
classification cs.CV cs.CL
keywords floor plan generationprogressive designdiffusion models3D indoor scene generationspatial priorsresidential layoutsketch conditioning
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.

Automated home layout tools usually demand a finished graph or text brief up front and often skip a real floor plan when building 3D interiors. The authors argue that real design starts from rough strokes and that a locked 2D plan is the only reliable spatial contract before furniture is placed. They train on partial sketches derived from 80K real plans, refine those sketches into precise vector plans with a coarse-to-fine diffusion model, then place furniture inside the verified rooms. The result is much closer to real plan distributions than prior 2D generators and scores higher with architects on spatial rationality than language-model-only 3D systems—even when the input sketch is only one-quarter complete.

Core claim

Design is progressive and the finished 2D floor plan is an irreplaceable spatial prior. A system that learns from partial sketches at every completeness level, sharpens them into geometrically valid vector plans, and only then furnishes inside those boundaries produces more faithful 2D layouts and more rational 3D scenes than methods that require fully specified inputs or that let language models invent layout and furniture together.

What carries the argument

SketchPlan plus PlanCraft-Diff: an automated replay of drafting that yields τ-labeled partial sketches, fed to a conditional diffusion model with modality-specific entry paths, two-stage coarse-to-fine training, and connectivity filtering; the verified plan then anchors PlanCraft-Agent’s constraint-based furniture placement.

Load-bearing premise

The automated step-by-step masking of walls then doors and windows on finished plans is assumed to stand in for the incomplete sketches architects actually draw, so scores at 25–100% completeness transfer to real progressive design.

What would settle it

Give practicing architects real mid-design hand sketches (not algorithmically masked finished plans), run PlanCraft and the strongest fully specified baselines on the same briefs, and check whether the 25%-sketch FID and expert rationality advantages still hold.

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

If this is right

  • Floor-plan generators can accept rough or partial strokes instead of complete adjacency graphs or detailed text.
  • A verified 2D plan should sit between language and 3D assembly; skipping it produces overlapping rooms and bad proportions.
  • Sparse sketches (~25% complete) already suffice to beat fully specified prior methods on standard layout metrics.
  • Coarse-to-fine diffusion plus connectivity filtering is a practical recipe for closed, aligned residential polygons.
  • Expert-rated spatial rationality in furnished multi-room scenes improves when furniture is solved inside locked room boundaries.

Where Pith is reading between the lines

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

  • The same progressive-sketch plus spatial-contract pattern may transfer to multi-floor buildings, offices, or site plans where language-only layout also fails.
  • If real architect sketches differ systematically from the wall-first masking order, a small human-sketch fine-tune set could be the highest-leverage next experiment.
  • Editability via language-updated vectors or CAD round-trip suggests a practical human-in-the-loop loop rather than one-shot generation.

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

4 major / 6 minor

Summary. The paper presents PlanCraft, a four-stage system that generates furnished 3D residential scenes from incomplete design sketches or natural-language programs. SketchPlan constructs partial-sketch training pairs from 80K RPLAN floor plans by masking structural elements under prerequisite constraints; PlanCraft-Diff is a modality-gated conditional DDPM with two-stage coarse-to-fine training and connectivity-based candidate filtering that refines incomplete inputs into complete floor plans; a rule-based post-processor vectorizes the raster output; and PlanCraft-Agent furnishes the verified vector plan using an LLM constraint generator plus a collision-aware solver. The authors report a 61.1% FID improvement over the best 2D baseline (HouseDiffusion), monotonic quality gains with sketch completeness (Fig. 3), ablations on backbone/training/filtering (Tab. 2), and a +15-point expert-rated Rationality advantage over Holodeck in a 26-participant architect study. The headline framing claims a 25%-complete sketch already outperforms all fully specified baselines.

Significance. If the results hold, this is a useful contribution: it is, to my knowledge, the first unified pipeline connecting progressive sketch completion to furnished 3D scene generation, and the "floor plan as spatial contract" argument is well-motivated and convincingly illustrated by the Holodeck/LLM failure cases in Fig. 6. Concrete strengths: a new 80K-plan partial-sketch dataset (SketchPlan) with a principled prerequisite-gated construction; a clean modality-gated DDPM design that shares parameters across two input types; ablations isolating backbone, curriculum, and filtering (Tab. 2); and an expert evaluation with 26 architecture-trained raters rather than crowdworkers. The graph-plus-block FID result (4.026 vs. 10.35 for HouseDiffusion) is large and, on its face, a genuine system-level result. The work is reproducible in outline: standard RPLAN data, standard DDPM/UNet, and full hyperparameters given.

major comments (4)
  1. [Fig. 3 / Abstract / Effect of Vector Completeness] The claim 'even at 25% completeness, PlanCraft surpasses all fully specified baselines' (Abstract; Experiments, 'Effect of Vector Completeness'; Fig. 3) compares conditions of unequal informativeness. Per the Experiments setup ('our automated pipeline progressively masks structural elements from complete plans'), the 25% vector input is an exact, noiseless subset of the ground-truth plan's walls/doors/windows at ground-truth coordinates — i.e., a quarter of the answer, unperturbed. Every Tab. 1 baseline instead receives only a graph or text prompt. FID/IoU/PSNR/SSIM are then computed against that same ground truth, so the comparison conflates input informativeness with progressive-design capability. This is the paper's most distinctive selling point and it is not apples-to-apples. Two fixes, either of which would suffice: (a) reframe the claim as an input-informativeness result and drop
  2. [Method (SDP) / Experiments setup] SDP simulates progressive drafting as exact subsets of the final vector plan with randomized order. A real architect's early sketch is rough, imprecise geometry encoding intent (approximate proportions, displaced walls), not a noiseless prefix of the final drawing. The paper motivates the entire system as 'architect-inspired' (Abstract; Introduction; Method, SDP paragraph), but no experiment tests robustness to geometric noise in the sketch (jittered vertices, approximate room extents), and no real or hand-drawn sketches are evaluated. As written, the Fig. 3 completeness curve characterizes performance on a synthetic proxy that may not transfer to the inputs the system claims to serve. At minimum, add a robustness study perturbing the partial vectors (vertex noise, room-block displacement at increasing sigma) and report metric degradation; a small qualitative study on actual architect sk
  3. [Tab. 2 / Method, Inference and Quality Filtering] The connectivity filter discards candidates whose floor-plan region is not a single connected component, and this alone moves FID from 11.674 to 4.026 — the single largest effect in the paper and most of the claimed 61.1% margin over HouseDiffusion (10.35). The manuscript does not state the evaluation protocol precisely: are Tab. 1 metrics computed only on the filtered subset of Ns=12 candidates? If so, FID is measured on a quality-curated sample, while baseline numbers (taken from prior publications or reproduced without filtering) are not given the same selection benefit, which would inflate the system-level comparison. The claim that filtering is 'a reliability filter instead of a hidden source of geometric improvement' is asserted but the IoU/PSNR/SSIM rows in Tab. 2 do move (0.498→0.510, etc.), so it is not purely distributional. Please (i) state exactly which samples enter each rep
  4. [Expert Evaluation on 3D Scene Generation] The user study (n=26) reports mean scores on three dimensions with a +15 Rationality gap, but the main text gives no variability or inferential statistics (no standard deviations, CIs, or significance tests), no statement on whether raters were blinded to system identity, and no discussion of ordering effects in the within-subjects design (each participant used both systems on 20 prompt-matched tasks). Given that PlanCraft's room boundaries are visually verifiable while Holodeck's failure modes are conspicuous (Fig. 6a), non-blind rating could inflate the gap. Please report per-dimension variance and a paired test, describe blinding and task randomization (even if in the supplement), and state whether participants had any affiliation with the project.
minor comments (6)
  1. [Tab. 1] Including Holodeck in Tab. 1 (a 2D floor-plan metric table) is questionable framing: Holodeck is a 3D system whose internal layout was never designed to be scored as a floor plan, and mapping its vectors to RPLAN labels stacks the deck. Consider moving this row to a footnote or clearly labeling it as diagnostic.
  2. [Tab. 2] Tab. 2's annotation '↓65.5%' for the full model is computed against the one-stage DDPM, while the text elsewhere quotes the two-stage gain as 15.5% (13.810→11.674) — consistent, but the mixed reference points will confuse readers; please make each percentage's baseline explicit.
  3. [Method / Implementation Details] Several free parameters are not ablated anywhere visible: the SDP wall-sampling probability λ, perturbation scales σ_blur and σ_block, the epoch split E1/E2=1000/500, and candidate count Ns=12. A short sensitivity table in the supplement would address reproducibility concerns.
  4. [Evaluation Metrics] All four Tab. 1 metrics are computed on 64×64 rasterized images. Since HouseDiffusion and HouseGAN++ are vector methods, rasterization resolution may differentially penalize them; please state how baseline outputs were rasterized and justify H=64, or report a vector-space metric (e.g., the authors' own metrics from Zeng et al. 2025, which is cited but not used).
  5. [Method, N-Graph construction] The N-graph node position p_i ~ U(bbox_i) is underspecified: is a fresh random position sampled per training pair, and does this stochasticity in the condition affect evaluation determinism? A sentence of clarification suffices.
  6. [Related Work] Related Work omits discussion of sketch-to-image/layout completion literature (e.g., scribble- or partial-layout-conditioned generation) outside residential floor plans; positioning against partial-conditioning methods in adjacent domains would clarify novelty of the modality-gated design.

Circularity Check

0 steps flagged

No significant circularity: empirical systems results on external FID/IoU/expert metrics, not inputs restated as predictions.

full rationale

PlanCraft is a conditional generative systems paper. Its load-bearing claims are measured against external benchmarks (FID, IoU, PSNR, SSIM on RPLAN-derived floor plans; expert Rationality/Practicality/Personalization scores vs Holodeck), not algebraic identities of fitted constants. SketchPlan builds supervised (partial, complete) pairs by masking or progressive sampling from real plans; the diffusion model must still synthesize the missing geometry, and success is scored against held-out-style complete targets—standard completion learning, not self-definition. The two-stage curriculum, modality-specific entry convolutions, and connectivity candidate filter are engineering choices whose effect is shown by ablation (Tab. 2); filtering discards disconnected samples but does not redefine the metric or force the reported layout quality by construction. Self-citations (e.g., Zeng et al., Yin et al.) appear as related work, baselines, or metric context, not as uniqueness theorems that forbid alternatives or smuggle the central result. The distinctive “25% sketch beats fully specified baselines” claim raises an evaluation-fairness / information-asymmetry issue (partial vectors are exact GT subsets while baselines get graph/text only), but that is not circularity under this pass: the completion is not equal to the input by definition, and the 61.1% FID headline itself is on the graph-plus-block condition. No step reduces a claimed prediction to its own fitted input or to a load-bearing unverified self-citation chain.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 3 invented entities

Load-bearing content is mostly engineering assumptions and dataset constructions rather than formal axioms. The central performance claims rest on RPLAN as a proxy for real residential design, on SDP as a proxy for human progressive sketching, on standard DDPM training, and on expert ratings as a proxy for spatial quality. Hyperparameters and filtering thresholds are chosen by the authors and materially affect FID.

free parameters (5)
  • two-stage epoch split E1/E2 = E1=1000, E2=500
    1000 coarse pretrain epochs + 500 fine epochs; ablation shows large FID effect vs one-stage.
  • coarse blur / block perturbation scales σ_blur, σ_block
    Gaussian perturbation scales for coarse pretraining and N-graph block construction; control how much global vs local signal the curriculum sees.
  • SDP wall sampling probability λ and completeness discretization c=⌊τ×100⌋ = τ reported at 25/50/75/100%
    Controls synthetic progressive sketch distribution and conditioning granularity; directly defines the 25/50/75/100% evaluation axis.
  • inference candidate count Ns and connectivity filter = Ns=12
    Ns=12 candidates with single-connected-component filtering; ablation drops FID from 11.674 to 4.026, so reported SOTA depends on this selection rule.
  • DDPM/UNet training hyperparameters = as listed in Implementation Details
    H=64, C=64, T=1000, β1=1e-4, βT=0.02, Adam lr 2e-4→2e-6, batch 64—standard but chosen settings that define the trained model.
axioms (6)
  • domain assumption Residential design is inherently progressive from incomplete sketches to refined plans, so partial-sketch conditioning is the right input modality.
    Stated as structural insight in Abstract/Introduction and used to justify SketchPlan rather than only full graphs/text.
  • domain assumption Once room boundaries, doors, and windows are fixed, furnishing reduces to bounded constraint satisfaction (the floor plan is a spatial contract).
    Core justification for PlanCraft-Agent versus end-to-end LLM layout (Abstract, Introduction, PlanCraft-Agent).
  • domain assumption RPLAN's ~80K real floor plans, after RPP vectorization, are a sufficient distribution for training and evaluating residential generators.
    All 2D training/eval and completeness curves are built from RPLAN (SketchPlan; Experiments).
  • standard math Standard conditional DDPM noise-prediction training yields samples from the data distribution useful for floor-plan images.
    Forward/reverse process and ℓ2 objective follow Ho et al. 2020 (PlanCraft-Diff equations 1–3).
  • ad hoc to paper Discarding candidates whose floor region is not a single connected component is a valid quality filter rather than a biased cherry-pick of the metric.
    Inference and Quality Filtering; ablation credits most of the final FID drop to this rule (Table 2).
  • domain assumption Expert ratings on Rationality/Practicality/Personalization from 26 architecture-trained participants measure true spatial quality of furnished scenes.
    3D evaluation relies entirely on this user study because automatic furnished-scene metrics are deemed inadequate (Expert Evaluation).
invented entities (3)
  • SketchPlan dataset (RPP + SDP + N-graph/perturbed blocks) no independent evidence
    purpose: Supply partial-sketch and graph-plus-block supervision at every completeness level from RPLAN.
    Constructed artifact; independent evidence is limited to the paper's own generation metrics, not an external standard benchmark release described in-text.
  • PlanCraft-Diff (modality-gated coarse-to-fine conditional DDPM) no independent evidence
    purpose: Progressively complete incomplete sketches or graph-block conditions into raster floor plans.
    System component; performance evidenced only via authors' RPLAN comparisons/ablations.
  • PlanCraft-Agent (constraint-generating LLM + collision-aware solver on verified vectors) no independent evidence
    purpose: Furnish multi-room 3D scenes inside fixed boundaries instead of free-form LLM layout.
    System component; +15 rationality vs Holodeck is author-run expert study, not external replication.

pith-pipeline@v1.2.0-grok45-kimik3 · 17238 in / 4007 out tokens · 84198 ms · 2026-07-30T20:55:50.546374+00:00 · methodology

0 comments
read the original abstract

Two structural insights have been overlooked in automated residential floor plan generation. First, design is inherently progressive. Architects begin with rough strokes and refine them over time, whereas existing methods typically require their conditioning representation to be fully specified before generation, a fundamental mismatch with how design actually works. Second, the 2D floor plan is not an optional intermediate but an irreplaceable spatial contract. Once room boundaries, doors, and windows are fixed, furnishing reduces from open-ended spatial reasoning to bounded constraint satisfaction. Bypassing this contract, as existing 3D systems do by delegating layout to language models, yields overlapping rooms and implausible proportions; directly calling general-purpose language models likewise produces geometrically invalid layouts. Guided by these insights, we present PlanCraft. SketchPlan supplies the missing training signal by replaying the architect's drawing process on 80K real floor plans, producing partial sketches at every completeness level. PlanCraft-Diff progressively sharpens an incomplete sketch into a geometrically precise, vectorizable floor plan through a coarse-to-fine strategy. With the spatial contract established, PlanCraft-Agent then furnishes the scene within well-defined room boundaries. Experiments show that PlanCraft achieves a 61.1\% lower FID than the best existing 2D method and surpasses existing 3D systems by 15 points in expert-rated spatial rationality, with a sketch at only 25\% completion already outperforming all fully specified baselines.

Figures

Figures reproduced from arXiv: 2607.23491 by Chaoyang Shi, Jing Zhong, Jun Yin, Maowei Jiang, Ng Cheuk Hei, Pengyu Zeng, Shuai Lu, Yuqin Dai, Yuxing Han, ZhanXiang Jin, Ziyang Han.

Figure 1
Figure 1. Figure 1: (a) Existing 2D methods take a graph or text prompt as input and use generative models to produce a 2D floor plan; they require a fully specified input and may still output an in￾valid layout (black circle).(b) Existing 3D methods take a text prompt and use an LLM agent to generate a 3D scene directly, without a dedicated data-driven floor-plan generation stage; this can produce incorrect room counts and o… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of PlanCraft. During training, SketchPlan constructs partial-sketch and graph-plus-block supervision for PlanCraft-Diff. At inference, a natural-language program is first converted by an LLM into a coarse graph-plus-block condition, while a partial vector sketch is directly rasterized as an alternative condition. PlanCraft-Diff then generates a complete floor plan, which is vectorized and furnishe… view at source ↗
Figure 3
Figure 3. Figure 3: Effect of vector completeness on generation quality. All four metrics improve monotonically, but at different rates: [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Representative outputs and failure cases across floor [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison from three perspectives. [PITH_FULL_IMAGE:figures/full_fig_p007_6.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

52 extracted references · 3 linked inside Pith

  1. [1]

    2011 , edition=

    Eastman, Chuck and Teicholz, Paul and Sacks, Rafael and Liston, Kathleen , title=. 2011 , edition=

    2011

  2. [2]

    Human-Computer Interaction -- INTERACT 2023 , pages=

    Supporting construction and architectural visualization through BIM and AR/VR: a systematic literature review , author=. Human-Computer Interaction -- INTERACT 2023 , pages=. 2023 , organization=

    2023

  3. [3]

    SN Computer Science , volume=

    From algorithms to architecture: Computational methods for house floorplan generation , author=. SN Computer Science , volume=. 2024 , publisher=

    2024

  4. [4]

    2020 , organization=

    Nauata, Nelson and Chang, Kai-Hung and Cheng, Chin-Yi and Mori, Greg and Furukawa, Yasutaka , booktitle=. 2020 , organization=

    2020

  5. [5]

    Nauata, Nelson and Hosseini, Sepidehsadat and Chang, Kai-Hung and Chu, Hang and Cheng, Chin-Yi and Furukawa, Yasutaka , booktitle=

  6. [6]

    Shabani, Mohammad Amin and Hosseini, Sepidehsadat and Furukawa, Yasutaka , booktitle=

  7. [7]

    Automation in Construction , volume=

    Residential floor plans: Multi-conditional automatic generation using diffusion models , author=. Automation in Construction , volume=. 2024 , publisher=

    2024

  8. [8]

    Buildings , volume=

    Comprehensive and dedicated metrics for evaluating AI-generated residential floor plans , author=. Buildings , volume=. 2025 , publisher=

    2025

  9. [9]

    Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Tell2design: A dataset for language-guided floor plan generation , author=. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  10. [10]

    Yin, Jun and Zeng, Pengyu and Sun, Haoyuan and Dai, Yuqin and Zheng, Han and Zhang, Miao and Zhang, Yachao and Lu, Shuai , booktitle=

  11. [11]

    Zeng, Pengyu and Yin, Jun and Zhang, Miao and Dai, Yuqin and Li, Jizhizi and Jin, Zhanxiang and Lu, Shuai , booktitle=

  12. [12]

    Proceedings of the 33rd ACM International Conference on Multimedia , pages=

    MRED-14: A Benchmark for Low-Energy Residential Floor Plan Generation with 14 Flexible Inputs , author=. Proceedings of the 33rd ACM International Conference on Multimedia , pages=

  13. [13]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    GreenPlanner: Practical Floorplan Layout Generation via an Energy-Aware and Function-Feasible Generative Framework , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  14. [14]

    Holodeck: Language Guided Generation of

    Yang, Yue and Sun, Fan-Yun and Weihs, Luca and VanderBilt, Eli and Herrasti, Alvaro and Han, Winson and Wu, Jiajun and Haber, Nick and Krishna, Ranjay and Liu, Lingjie and Callison-Burch, Chris and Yatskar, Mark and Kembhavi, Aniruddha and Clark, Christopher , booktitle=. Holodeck: Language Guided Generation of

  15. [15]

    Deitke, Matt and Vander Bilt, Eli and Herrasti, Alvaro and Weihs, Luca and Salvador, Jordi and Ehsani, Kiana and Han, Winson and Kolve, Eric and Farhadi, Ali and Kembhavi, Aniruddha and others , booktitle=

  16. [16]

    Yang, Yixuan and Lu, Junru and Zhao, Zixiang and Luo, Zhen and Yu, James JQ and Sanchez, Victor and Zheng, Feng , journal=

  17. [17]

    Lin, Chenguo and Mu, Yadong , booktitle=

  18. [18]

    Yang, Yixuan and Luo, Zhen and Ding, Tongsheng and Lu, Junru and Gao, Mingqi and Yang, Jinyu and Sanchez, Victor and Zheng, Feng , booktitle=

  19. [19]

    2020 , publisher=

    Hu, Ruizhen and Huang, Zeyu and Tang, Yuhan and Van Kaick, Oliver and Zhang, Hao and Huang, Hui , journal=. 2020 , publisher=

    2020

  20. [20]

    2022 , publisher=

    Sun, Jiahui and Wu, Wenming and Liu, Ligang and Min, Wenjie and Zhang, Gaofeng and Zheng, Liping , journal=. 2022 , publisher=

    2022

  21. [21]

    ACM Transactions on Graphics (TOG) , volume=

    Data-Driven Interior Plan Generation for Residential Buildings , author=. ACM Transactions on Graphics (TOG) , volume=. 2019 , publisher=

    2019

  22. [22]

    Advances in neural information processing systems , volume=

    Denoising Diffusion Probabilistic Models , author=. Advances in neural information processing systems , volume=

  23. [23]

    2015 , organization=

    Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas , booktitle=. 2015 , organization=

    2015

  24. [24]

    International Conference on Learning Representations (

    Adam: A Method for Stochastic Optimization , author=. International Conference on Learning Representations (. 2015 , url=

    2015

  25. [25]

    International Conference on Learning Representations (

    Flow Matching for Generative Modeling , author=. International Conference on Learning Representations (. 2023 , url=

    2023

  26. [26]

    2025 , eprint=

    Yang, An and Li, Anfeng and Yang, Baosong and Zhang, Beichen and Hui, Binyuan and Zheng, Bo and Yu, Bowen and Gao, Chang and Huang, Chengen and Lv, Chenxu and Zheng, Chujie and Liu, Dayiheng and Zhou, Fan and Huang, Fei and Hu, Feng and Ge, Hao and Wei, Haoran and Lin, Huan and Tang, Jialong and Yang, Jian and Tu, Jianhong and Zhang, Jianwei and Yang, Jia...

    2025

  27. [27]

    2025 , howpublished =

    Doubao:. 2025 , howpublished =

    2025

  28. [28]

    Reimers, Nils and Gurevych, Iryna , booktitle=

  29. [29]

    Objaverse: A Universe of Annotated

    Deitke, Matt and Schwenk, Dustin and Salvador, Jordi and Weihs, Luca and Michel, Oscar and VanderBilt, Eli and Schmidt, Ludwig and Ehsani, Kiana and Kembhavi, Aniruddha and Farhadi, Ali , booktitle=. Objaverse: A Universe of Annotated

  30. [30]

    2017 , eprint=

    Kolve, Eric and Mottaghi, Roozbeh and Han, Winson and VanderBilt, Eli and Weihs, Luca and Herrasti, Alvaro and Deitke, Matt and Ehsani, Kiana and Gordon, Daniel and Zhu, Yuke and Kembhavi, Aniruddha and Gupta, Abhinav and Farhadi, Ali , journal=. 2017 , eprint=

    2017

  31. [31]

    Architectural intelligence: Selected papers from the 1st international conference on computational design and robotic fabrication (CDRF 2019) , pages=

    Archigan: Artificial intelligence x architecture , author=. Architectural intelligence: Selected papers from the 1st international conference on computational design and robotic fabrication (CDRF 2019) , pages=. 2020 , organization=

    2019

  32. [32]

    Proceedings of the 2025 International Conference on Multimedia Retrieval , pages=

    Floorplan-Diffusion: Automatic Floor Plan Generation via Pre-trained Large Latent Diffusion Model , author=. Proceedings of the 2025 International Conference on Multimedia Retrieval , pages=

    2025

  33. [33]

    Feng, Weixi and Zhu, Wanrong and Fu, Tsu-Jui and Jampani, Varun and Akula, Arjun and He, Xuehai and Basu, S and Wang, Xin Eric and Wang, William Yang , booktitle=

  34. [34]

    arXiv preprint arXiv:2507.06261 , year=

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities , author=. arXiv preprint arXiv:2507.06261 , year=. 2507.06261 , archivePrefix=

    Pith/arXiv arXiv

  35. [35]

    Heusel, Martin and Ramsauer, Hubert and Unterthiner, Thomas and Nessler, Bernhard and Hochreiter, Sepp , booktitle=

  36. [36]

    and Perelman, Adam and Ramesh, Aditya and Clark, Aidan and Ostrow, AJ and Welihinda, Akila and Hayes, Alan and Radford, Alec and others , journal=

    Hurst, Aaron and Lerer, Adam and Goucher, Adam P. and Perelman, Adam and Ramesh, Aditya and Clark, Aidan and Ostrow, AJ and Welihinda, Akila and Hayes, Alan and Radford, Alec and others , journal=. 2024 , eprint=

    2024

  37. [37]

    Proceedings of the 38th International Conference on Machine Learning , pages=

    Learning Transferable Visual Models From Natural Language Supervision , author=. Proceedings of the 38th International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  38. [38]

    Advances in Neural Information Processing Systems , volume=

    Attention Is All You Need , author=. Advances in Neural Information Processing Systems , volume=

  39. [39]

    SSIM , author=

    Image quality metrics: PSNR vs. SSIM , author=. 2010 20th international conference on pattern recognition , pages=. 2010 , organization=

    2010

  40. [40]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Generalized Intersection over Union: A Metric and A Loss for Bounding Box Regression , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  41. [41]

    IEEE transactions on image processing , volume=

    Image quality assessment: from error visibility to structural similarity , author=. IEEE transactions on image processing , volume=. 2004 , publisher=

    2004

  42. [42]

    2018 , edition=

    BIM Handbook: A Guide to Building Information Modeling for Owners, Designers, Engineers, Contractors, and Facility Managers , author=. 2018 , edition=

    2018

  43. [43]

    2024 , eprint=

    I-Design: Personalized LLM Interior Designer , author=. 2024 , eprint=

    2024

  44. [44]

    arXiv preprint arXiv:2403.05530 , year=

    Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context , author=. arXiv preprint arXiv:2403.05530 , year=. 2403.05530 , archivePrefix=

    Pith/arXiv arXiv

  45. [45]

    International Conference on Learning Representations (ICLR) , year=

    SGDR: Stochastic Gradient Descent with Warm Restarts , author=. International Conference on Learning Representations (ICLR) , year=

  46. [46]

    Advances in Neural Information Processing Systems , volume=

    Mean Flows for One-step Generative Modeling , author=. Advances in Neural Information Processing Systems , volume=

  47. [47]

    2026 , howpublished =

    2026

  48. [48]

    2025 , howpublished =

    2025

  49. [49]

    2607.00248 , archivePrefix =

    arXiv preprint arXiv:2607.00248 , year =. 2607.00248 , archivePrefix =

    Pith/arXiv arXiv

  50. [50]

    2026 , eprint=

    Xia, Hongchi and Li, Xuan and Li, Zhaoshuo and Ma, Qianli and Xu, Jiashu and Liu, Ming-Yu and Cui, Yin and Lin, Tsung-Yi and Ma, Wei-Chiu and Wang, Shenlong and Song, Shuran and Wei, Fangyin , booktitle=. 2026 , eprint=

    2026

  51. [51]

    2026 , eprint=

    Che, Lirong and Wen, Shuo and Huang, Shan and Wang, Chuang and Yang, Yuzhe and Dudek, Gregory and Wang, Xueqian and Su, Jian , booktitle=. 2026 , eprint=

    2026

  52. [52]

    2026 , eprint=

    Wang, Dilin and Jung, Hyunyoung and Monnier, Tom and Sohn, Kihyuk and Zou, Chuhang and Xiang, Xiaoyu and Yeh, Yu-Ying and Liu, Di and Huang, Zixuan and Nguyen-Phuoc, Thu and Fan, Yuchen and Oprea, Sergiu and Wang, Ziyan and Shapovalov, Roman and Sarafianos, Nikolaos and Groueix, Thibault and Toisoul, Antoine and Dhar, Prithviraj and Chu, Xiao and Chen, Mi...

    2026