Pith. sign in

REVIEW 1 major objections 5 minor 41 references

LayoutGKN computes floor-plan similarity by learning independent node embeddings and scoring them with a differentiable graph kernel, matching or beating the slower graph matching network while cutting inference time by roughly 20x.

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 ↗

LayoutGKN uses a differentiable path-based graph kernel over learned room embeddings to rank floor plans, matching LayoutGMN's accuracy at about 20x lower inference cost.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Real speedup, shaky P@k protocol: the ranking claim needs a proper evaluation, but the triplet accuracy and efficiency hold up. the 1 major comments →

arxiv 2509.03737 v1 pith:EVFGHHXR submitted 2025-09-03 cs.CV

LayoutGKN: Graph Similarity Learning of Floor Plans

Casper van Engelenburg , Jan van Gemert , Seyran Khademi This is my paper
classification cs.CV
keywords floor plan retrievalgraph similarity learninggraph matching networksgraph kernelgraph edit distancejoint embedding architecturezero-shot generalizationdifferentiable kernel
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.

The reading

Floor-plan similarity is usually computed with graph matching networks, which interleave cross-graph node interactions throughout the encoder and therefore cannot precompute embeddings, making them slow. LayoutGKN instead learns node embeddings independently and moves all cross-graph node-level interactions into a differentiable path-based graph kernel applied at the end of the pipeline. This preserves the accuracy of node-level correspondence while letting gallery embeddings be precomputed offline. On the RPLAN and MSD benchmarks, the method matches or beats the LayoutGMN baseline on retrieval precision while reducing inference time by about an order of magnitude (roughly 20x per 10K pairs), and it reports stronger zero-shot transfer to MSD.

Core claim

The central claim is that explicit cross-graph node-level interactions inside graph encoders are not necessary for effective graph similarity; they can be postponed to a differentiable kernel that scores the final independently learned node embeddings. LayoutGKN embeds each floor plan with a shared message-passing network, then computes similarity with a normalized GraphHopper kernel over the node embeddings. The kernel weighs node pairs by the overlap of their shortest-path histograms, so topological correspondences are still captured at scoring time, but the two graphs never interact during encoding. Because embeddings are computed independently, they can be precomputed for a gallery, whic

What carries the argument

The load-bearing component is the differentiable GraphHopper path-based graph kernel applied over learned node embeddings. Each node is endowed with a shortest-path histogram matrix M(u) that counts how often the node appears at each position along shortest paths of each length; the kernel then computes kG(G1, G2) = Σ_{u∈V1} Σ_{v∈V2} ⟨M(u), M(v)⟩ · k_node(h_u, h_v), where k_node is a Gaussian kernel on the learned embeddings. This placement lets the encoder process each graph in isolation while the similarity function still performs a form of node-level matching. A normalized version of the kernel is used in a triplet loss, and because both the kernel and the embeddings are differentiable, t

Load-bearing premise

The whole evaluation rests on the premise that the normalized graph edit distance (sGED) used as ground truth agrees with how humans perceive floor-plan spatial similarity; if that link is wrong, the reported ranking quality and zero-shot gains lose their meaning, and the user study backing it up is small (50 queries) with acknowledged disagreement.

What would settle it

Conduct a user study with a few hundred floor-plan queries, asking architects to rank near-duplicate plans; if rankings correlate more strongly with mean intersection-over-union than with normalized graph edit distance, the sGED training target that LayoutGKN optimizes would be invalidated, and the reported precision numbers would no longer reflect human-perceived similarity.

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

If this is right

  • Real-time retrieval over tens of thousands of floor plans becomes practical: per-query scoring costs about 1–2 seconds per 10K pairs because gallery embeddings are precomputed.
  • The node embeddings can be much smaller without a large accuracy drop, since the kernel, rather than the embedding dimension, carries the topological information.
  • The zero-shot result on MSD indicates that the kernel imposes a structural prior that transfers across floor-plan datasets, although the paper notes part of the gain may come from dataset composition.
  • Training and evaluation under normalized graph edit distance, rather than IoU, aligns better with human ranking, at least in the reported user study.

Where Pith is reading between the lines

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

  • The same postpone-and-kernel strategy could generalize to other graph comparison tasks, such as molecular similarity or program syntax trees, where cross-graph attention is the computational bottleneck.
  • Because gallery embeddings are precomputed, LayoutGKN could be combined with approximate nearest-neighbor indexing to scale to millions of floor plans without altering the training objective.
  • A testable extension is replacing the fixed shortest-path histogram with a learned topology descriptor, which might adapt the kernel to datasets with different edge semantics.
  • The method makes incremental database updates cheap: adding a new floor plan only requires computing its own embedding and kernel self-normalizer, not pairwise interactions with every existing plan.
Share X Bluesky LinkedIn Reddit HN

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

1 major / 5 minor

Summary. The paper proposes LayoutGKN, a graph-kernel network for floor-plan similarity retrieval. It replaces the cross-graph node-interaction layers of graph matching networks (GMNs) with a differentiable GraphHopper kernel computed on independently learned node embeddings, so that node embeddings can be precomputed offline. Training uses triplets mined with a normalized graph-edit-distance target (sGED), and evaluation is on RPLAN with zero-shot transfer to MSD, comparing triplet accuracy, P@5/P@10, and inference time against LayoutGMN, LayoutGEN, and a non-learning graph kernel. The paper reports comparable or better ranking accuracy than LayoutGMN (RPLAN P@5 0.623 vs 0.616, P@10 0.683 vs 0.675) with roughly a 20x speedup per 10K pairs, and stronger zero-shot P@k on MSD.

Significance. The architectural insight is clear and plausible: cross-graph interactions can be postponed to the similarity function without sacrificing accuracy, enabling precomputation and much faster retrieval. The method is simple, the equations are internally consistent, the loss is differentiable, and the authors provide open code and data. If the ranking evaluation is repaired, this would be a useful contribution to floor-plan retrieval and to graph-similarity learning more broadly. The speed advantage over GMNs is real and well motivated, since GMN embeddings cannot be precomputed. The paper also includes a user study connecting sGED to human judgment, although its scale is limited.

major comments (1)
  1. [Eq. (9) and GED computation] The definition of sGED assumes an exact GED value, but GED is NP-hard and the paper does not state how it is computed (exact algorithm, approximation, or upper bound) or give the maximum graph size. Since sGED is used both for triplet mining and for ground-truth evaluation, an uncontrolled approximation in Eq. (9) could affect all results. Please specify the GED solver and any limits on graph size.
minor comments (5)
  1. [Sec. 5.1 / Table 1] The text says 'P@5 and 20' but Table 1 reports P@5 and P@10. Please correct the typo.
  2. [Sec. 5.1] The term 'triplet accuracy' is used without a definition. Please state whether it is the fraction of triplets where the anchor-positive distance is smaller than anchor-negative, or some other criterion.
  3. [Sec. 4.2] The triplet-mining ranges (positive sGED in (0.6,0.9), negative ratio in (0.7,0.9)) are introduced as fixed heuristics. A sentence on the sensitivity of results to these thresholds, or a reference to prior use, would help.
  4. [Sec. 3.1] The shape feature vector is written as s = [cx; cy; w; h; sqrt(a); p/4] in Sec. 3.1 but in the supplementary it is [c(u)_x, c(u)_y, w, h, sqrt(a), p/4]^T. The text in Sec. 3.1 also uses 'lx' and 'ly' inconsistently. Please align notation.
  5. [Fig. 4 caption / Sec. 5.2] The qualitative example is helpful, but the caption should specify whether the shown rankings are from the top-50 protocol or from full-corpus ranking. This matters for interpreting the gray-background items.

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained, with minor self-referential evaluation choices.

full rationale

The paper's derivation chain is self-contained. LayoutGKN's similarity function (Eq. 5) is the fixed GraphHopper kernel from external prior work [6]; node embeddings are learned by a GNN whose parameters are optimized via triplet loss (Eq. 7) against the external sGED ground truth on held-out RPLAN test pairs. No equation reduces to its input: the kernel weights are not fit to the evaluation metric, and sGED is computed independently of the model. The zero-shot MSD evaluation uses an externally published (though author-contributed) dataset and involves no training on MSD. The only mild self-reference is that the same sGED metric defines training triplets and evaluation ground truth; this is standard supervised learning, not a tautology. The supplementary 'Precision' section defines P@k in a non-standard way (top-50 model candidates re-ranked by ground truth); this is a methodological limitation affecting retrieval comparisons, but it is not circular because the ground-truth sGED is external and the model's ranking is compared against it. Self-citations [31] and [32] are used for dataset cleaning and the MSD benchmark; they are not load-bearing for the method's mathematical derivation. Overall, no circular step is present; the central claims retain independent content.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The method introduces no new physical or computational entities. It relies on a fixed graph kernel from prior literature, learned node embeddings, and a hand-designed triplet mining strategy. The free parameters are conventional hyperparameters; the main burden is the validity of the sGED ground truth and the truncated path histogram representation.

free parameters (3)
  • Gaussian kernel width parameter mu = not reported; stated proportional to 1/d
    In Eq. 6, k_node = exp(-mu ||h_u - h_v||^2). mu controls the sharpness of the learned node similarity and is a hyperparameter selected by grid search (suppl. Hyperparameters). The exact constant of proportionality is not given.
  • Triplet margin m = searched in {5e-4, 1e-4, ..., 5e-1, 1}
    In Eq. 7, m is the logistic margin of the triplet loss, chosen by hyperparameter search to maximize validation accuracy.
  • Hidden dimension d and number of layers L = d in {8,16,32,64}, L in {2,3,4,5}; final values not explicitly stated
    Architecture capacities selected through hyperparameter analysis, with Fig. 3 showing the effect on accuracy. These are standard model selection choices, not fitted constants in a derivation.
axioms (4)
  • domain assumption sGED (normalized graph edit distance, Eq. 9) is a valid proxy for human-judged floor plan spatial similarity.
    Invoked as the training objective and evaluation ground truth (Sec 3.2, 5.2). Support is a user study of 50 queries with acknowledged inter-annotator disagreement; the correlation is described qualitatively rather than with full statistics.
  • domain assumption The GraphHopper kernel (Eq. 5) with shortest-path histograms truncated at delta=4 captures the topology relevant for GED-based floor plan similarity.
    Introduced in Sec 4.1 and Sec 3.1 ('we set delta = 4'). If floor plan graph diameters exceed 4, path information is truncated; the kernel's fixed topological prior cannot be overridden by the learned embeddings beyond the Gaussian term.
  • ad hoc to paper Triplet mining heuristics (positive with 0.9 > sGED > 0.6, negative with ratio in (0.7, 0.9)) produce informative training signal.
    Described in Sec 4.2. These thresholds are not derived or validated, but all methods use the same mining strategy so the comparison is fair.
  • standard math Standard GNN message passing and GRU update (Eqs. 3-4) from [15] are used without modification for intra-graph encoding.
    Sec 4.1; standard practice in the graph neural network literature, not a source of new error.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of LayoutGKN: Graph Similarity Learning of Floor Plans." pith.science (2026). https://pith.science/paper/EVFGHHXR

@misc{pith2026250903737,
  author       = {Pith},
  title        = {Pith review of: LayoutGKN: Graph Similarity Learning of Floor Plans},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EVFGHHXR}},
  note         = {Machine review of arXiv:2509.03737}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Floor plans depict building layouts and are often represented as graphs to capture the underlying spatial relationships. Comparison of these graphs is critical for applications like search, clustering, and data visualization. The most successful methods to compare graphs \ie, graph matching networks, rely on costly intermediate cross-graph node-level interactions, therefore being slow in inference time. We introduce \textbf{LayoutGKN}, a more efficient approach that postpones the cross-graph node-level interactions to the end of the joint embedding architecture. We do so by using a differentiable graph kernel as a distance function on the final learned node-level embeddings. We show that LayoutGKN computes similarity comparably or better than graph matching networks while significantly increasing the speed. \href{https://github.com/caspervanengelenburg/LayoutGKN}{Code and data} are open.

Figures

Figures reproduced from arXiv: 2509.03737 by Casper van Engelenburg, Jan van Gemert, Seyran Khademi.

Figure 1
Figure 1. Figure 1: Joint embeddings architectures for graph similarity learning. (Left) Graph embeddings networks (GENs) independently embed graphs G1 and G2 in a vector space using parameter-shared graph networks (gray box) followed by pooling (orange box). The similarity s is computed based on the vector embeddings. (Middle) In addition to GENs, graph matching networks (GMNs) model cross￾graph node-level interactions (blue… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of LayoutGKN. (A. graph extraction) Semantic images from RPLAN are converted into richly-attributed access graphs. The floor plans’ geometries (i.e., the rooms represented as polygons) are centered at (0, 0) and scaled to fit within the unit square box. The unit box amounts to 20 x 20 meters in reality (i.e., 0.1 equals 1 meter). The color indicates the room’s semantic category (e.g., dark orange … view at source ↗
Figure 3
Figure 3. Figure 3: (Left and middle) The effect of the number of learnable parameters on the triplet accuracy. We vary the number of parameters by changing the hidden node dimension and number of graph convolutional layers. The hidden node dimension has a profound effect on the accuracy: the smaller the node dimension, the more a decrease in accuracy. The rate at which the accuracy drops for decreasing node dimension is lowe… view at source ↗
Figure 4
Figure 4. Figure 4: Examples of ranking on RPLAN. (A) Given a query floor plan on the left, the top row shows the ground truth (GT) ranking. The value on top of the floor plan indicates the similarity in terms of sGED. Since the GT scores are quite often equivalent for different retrievals, we color-coded them: the same color means the same GT score. Rows 2 to 4 depict the rankings based on GEN, GMN, and GKN. On top of the fl… view at source ↗
Figure 5
Figure 5. Figure 5: Floor plan datasets and representation. Data Datasets: RPLAN and MSD We use RPLAN [35] and MSD [32]. RPLAN contains 88K+ floor plans and covers Asian residential apartments. As shown by [31], RPLAN contains a substantial amount of near-duplicates as well as floor plans that are not entirely connected (i.e., floor plans that contain rooms or room constellations that are disconnected from the rest). We remov… view at source ↗
Figure 6
Figure 6. Figure 6: Setup of user study in Miro. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: User study results. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_7.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

41 extracted references · 40 canonical work pages

  1. [1]

    A Pattern Language: Towns, Buildings, Construction

    Christopher Alexander, Ishikawa Sara, and Silverstein Mur- ray. A Pattern Language: Towns, Buildings, Construction . Oxford University Press, 1977. 2, 3

    work page 1977

  2. [2]

    SimGNN: A Neural Network Approach to Fast Graph Similarity Computation

    Yunsheng Bai, Hao Ding, Song Bian, Ting Chen, Yizhou Sun, and Wei Wang. SimGNN: A Neural Network Approach to Fast Graph Similarity Computation. In Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining, 2019. 2

    work page 2019

  3. [3]

    Layout Representation Learning with Spatial and Structural Hierarchies

    Yue Bai, Dipu Manandhar, Zhaowen Wang, John Collo- mosse, and Yun Fu. Layout Representation Learning with Spatial and Structural Hierarchies. In AAAI, 2023. 2

    work page 2023

  4. [4]

    Signature Verification using a ”Siamese” Time Delay Neural Network

    Jane Bromley, Isabelle Guyon, Yann LeCun, Eduard S¨ackinger, and Roopak Shah. Signature Verification using a ”Siamese” Time Delay Neural Network. In NeurIPS, 1993. 2

    work page 1993

  5. [5]

    A Simple Framework for Contrastive Learn- ing of Visual Representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton. A Simple Framework for Contrastive Learn- ing of Visual Representations. ICML, 2020. 2

    work page 2020

  6. [6]

    Scalable Kernels for Graphs with Continuous Attributes

    Aasa Feragen, Niklas Kasenburg, Jens Petersen, Marleen de Bruijne, and Karsten Borgwardt. Scalable Kernels for Graphs with Continuous Attributes. In NeurIPS, 2013. 2, 3, 5, 6

    work page 2013

  7. [7]

    Char- acterizing Structural Relationships in Scenes Using Graph Kernels

    Matthew Fisher, Manolis Savva, and Pat Hanrahan. Char- acterizing Structural Relationships in Scenes Using Graph Kernels. In ACM SIGGRAPH, 2011. 2

    work page 2011

  8. [8]

    COHO: Context-Sensitive City- Scale Hierarchical Urban Layout Generation

    Liu He and Daniel Aliaga. COHO: Context-Sensitive City- Scale Hierarchical Urban Layout Generation. In ECCV,

    work page

  9. [9]

    Segment Anything Model for Road Network Graph Extraction

    Congrui Hetang, Haoru Xue, Cindy Le, Tianwei Yue, Wen- ping Wang, and Yihui He. Segment Anything Model for Road Network Graph Extraction. In CVPRw, 2024. 1

    work page 2024

  10. [10]

    Space Syn- tax

    B Hillier, A Leaman, P Stansall, and M Bedford. Space Syn- tax. Environment and Planning B: Planning and Design ,

    work page

  11. [11]

    Deep Metric Learning Using Triplet Network

    Elad Hoffer and Nir Ailon. Deep Metric Learning Using Triplet Network. In Similarity-Based Pattern Recognition ,

    work page

  12. [12]

    SHRAG: Semantic Hierarchical Graph for Floorplan Representation

    Jiongchao Jin, Zhou Xue, and Biao Leng. SHRAG: Semantic Hierarchical Graph for Floorplan Representation. InInterna- tional Conference on 3D Vision, 2022. 2, 3, 5

    work page 2022

  13. [13]

    An Interactive Floor Plan Image Retrieval Framework Based on Structural Features

    Rasika Khade, Krupa Jariwala, and Chiranjoy Chattopad- hyay. An Interactive Floor Plan Image Retrieval Framework Based on Structural Features. Arabian Journal for Science and Engineering, 2023. 2

    work page 2023

  14. [14]

    Kriege, Fredrik D

    Nils M. Kriege, Fredrik D. Johansson, and Christopher Mor- ris. A Survey on Graph Kernels. Applied Network Science,

    work page

  15. [15]

    Graph Matching Networks for Learning the Similarity of Graph Structured Objects

    Yujia Li, Chenjie Gu, Thomas Dullien, Oriol Vinyals, and Pushmeet Kohli. Graph Matching Networks for Learning the Similarity of Graph Structured Objects. In ICML, 2019. arXiv:1904.12787. 2, 4, 5

    Pith/arXiv arXiv 2019

  16. [16]

    Decoupled Weight Decay Regularization

    Ilya Loshchilov and Frank Hutter. Decoupled Weight Decay Regularization. In ICLR, 2019. 5

    work page 2019

  17. [17]

    Ahmed, Theodore L

    Guixiang Ma, Nesreen K. Ahmed, Theodore L. Willke, and Philip S. Yu. Deep Graph Similarity Learning: A Survey,

    work page

  18. [18]

    Learning Structural Similarity of User Interface Layouts Using Graph Networks

    Dipu Manandhar, Dan Ruta, and John Collomosse. Learning Structural Similarity of User Interface Layouts Using Graph Networks. In ECCV, 2020. 2, 3, 5

    work page 2020

  19. [19]

    House-GAN: Relational Generative Adversarial Networks for Graph-Constrained House Layout Generation

    Nelson Nauata, Kai-Hung Chang, Chin-Yi Cheng, Greg Mori, and Yasutaka Furukawa. House-GAN: Relational Generative Adversarial Networks for Graph-Constrained House Layout Generation. In ECCV, 2020. 1, 2, 3

    work page 2020

  20. [20]

    Floor Plan Recommendation System using Graph Neural Network with Spatial Relationship Dataset.Journal of Build- ing Engineering, 2023

    Hyejin Park, Hyegyo Suh, Jaeil Kim, and Seungyeon Choo. Floor Plan Recommendation System using Graph Neural Network with Spatial Relationship Dataset.Journal of Build- ing Engineering, 2023. 2

    work page 2023

  21. [21]

    LayoutGMN: Neural Graph Match- ing for Structural Layout Similarity

    Akshay Gadi Patil, Manyi Li, Matthew Fisher, Manolis Savva, and Hao Zhang. LayoutGMN: Neural Graph Match- ing for Structural Layout Similarity. In CVPR, 2021. 1, 2, 3, 5, 6

    work page 2021

  22. [22]

    Pizarro, Nancy Hitschfeld, Ivan Sipiran, and Jose M

    Pablo N. Pizarro, Nancy Hitschfeld, Ivan Sipiran, and Jose M. Saavedra. Automatic Floor Plan Analysis and Recognition. Automation in Construction, 140, 2022. 1

    work page 2022

  23. [23]

    Learning Graph Edit Distance by Graph Neural Networks

    Pau Riba, Andreas Fischer, Josep Llad ´os, and Alicia Forn´es. Learning Graph Edit Distance by Graph Neural Networks. Pattern Recognition, 2020. 2

    work page 2020

  24. [24]

    Semantic Pattern-based Retrieval of Architectural Floor Plans with Case-based and Graph-based Searching Techniques and their Evaluation and Visualization

    Qamer Uddin Sabri, Johannes Bayer, Viktor Ayzenshtadt, Syed Saqib Bukhari, Klaus-Dieter Althoff, and Andreas Dengel. Semantic Pattern-based Retrieval of Architectural Floor Plans with Case-based and Graph-based Searching Techniques and their Evaluation and Visualization. In Pro- ceedings of the 6th International Conference on Pattern Recognition Applicati...

    work page 2017

  25. [25]

    A Distance Measure Be- tween Attributed Relational Graphs for Pattern Recognition

    Alberto Sanfeliu and King-Sun Fu. A Distance Measure Be- tween Attributed Relational Graphs for Pattern Recognition. IEEE Transactions on Systems, Man, and Cybernetics , 13,

    work page

  26. [26]

    High-level Feature Aggregation for Fine-Grained Architectural Floor Plan Retrieval

    Divya Sharma and Chiranjoy Chattopadhyay. High-level Feature Aggregation for Fine-Grained Architectural Floor Plan Retrieval. IET Computer Vision, 2018. 2

    work page 2018

  27. [27]

    Borgwardt

    Nino Shervashidze, Pascal Schweitzer, Erik Jan van Leeuwen, Kurt Mehlhorn, and Karsten M. Borgwardt. Weisfeiler-Lehman Graph Kernels. Journal of Machine Learning Research, 2011. 2

    work page 2011

  28. [28]

    Matthias Standfest, Michael Franzen, Yvonne Schr ¨oder, Luis Gonzalez Medina, Yarilo Villanueva Hernandez, Jan Hendrik Buck, Yen-Ling Tan, Milena Niedzwiecka, and Rachele Colmegna. Swiss Dwellings: A Large Dataset of Apartment Models Including Aggregated Geolocation-based Simulation Results Covering Viewshed, Natural Light, Traf- fic Noise, Centrality and...

    work page arXiv 2022

  29. [29]

    Architectural Morphology: An Introduc- tion to the Geometry of Building Plans

    Philip Steadman. Architectural Morphology: An Introduc- tion to the Geometry of Building Plans. Pion, 1983. 2, 3

    work page 1983

  30. [30]

    Similar Floor Plan Retrieval featuring Multi- Task Learning of Layout Type Classification and Room Pres- ence Prediction

    Yuki Takada, Naoto Inoue, Toshihiko Yamasaki, and Kiy- oharu Aizawa. Similar Floor Plan Retrieval featuring Multi- Task Learning of Layout Type Classification and Room Pres- ence Prediction. In ICCE, 2018. 2 8

    work page 2018

  31. [31]

    SSIG: A Visually-Guided Graph Edit Distance for Floor Plan Similarity

    Casper van Engelenburg, Seyran Khademi, and Jan van Gemert. SSIG: A Visually-Guided Graph Edit Distance for Floor Plan Similarity. In ICCVw, 2023. 2, 10

    work page 2023

  32. [32]

    MSD: A Benchmark Dataset for Floor Plan Generation of Building Complexes

    Casper van Engelenburg, Fatemeh Mostafavi, Jan van Gemert, and Seyran Khademi. MSD: A Benchmark Dataset for Floor Plan Generation of Building Complexes. InECCV,

    work page

  33. [33]

    Graph At- tention Networks

    Petar Veli ˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li`o, and Yoshua Bengio. Graph At- tention Networks. In ICLR, 2018. 6

    work page 2018

  34. [34]

    The Room Connectivity Graph: Shape Retrieval in the Architectural Domain

    Raoul Wessel, Ina Bl ¨umel, and Reinhard Klein. The Room Connectivity Graph: Shape Retrieval in the Architectural Domain. In The 16-th International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, 2008. 2

    work page 2008

  35. [35]

    Data-driven Interior Plan Genera- tion for Residential Buildings

    Wenming Wu, Xiao-Ming Fu, Rui Tang, Yuhan Wang, Yu- Hao Qi, and Ligang Liu. Data-driven Interior Plan Genera- tion for Residential Buildings. ACM ToG, 38, 2019. 5, 10

    work page 2019

  36. [36]

    Cross-Domain Doc- ument Layout Analysis via Unsupervised Document Style Guide

    Xingjiao Wu, Luwei Xiao, Xiangcheng Du, Yingbin Zheng, Xin Li, Tianlong Ma, and Liang He. Cross-Domain Doc- ument Layout Analysis via Unsupervised Document Style Guide. Expert Syst. Appl., 2022. 2

    work page 2022

  37. [37]

    How Powerful are Graph Neural Networks? In ICLR, 2019

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How Powerful are Graph Neural Networks? In ICLR, 2019. 6

    work page 2019

  38. [38]

    Distance Metric Learning: A Comprehensive Survey

    Liu Yang and Rong Jin. Distance Metric Learning: A Comprehensive Survey. https://www.cs.cmu.edu/ ˜liuy/frame_survey_v2.pdf, 2006. 3

    work page 2006

  39. [39]

    Zhiping Zeng, Anthony K. H. Tung, Jianyong Wang, Jianhua Feng, and Lizhu Zhou. Comparing Stars: On approximating Graph Edit Distance. Proc. VLDB Endow., 2009. 2

    work page 2009

  40. [40]

    GraSP: Sim- ple yet Effective Graph Similarity Predictions

    Haoran Zheng, Jieming Shi, and Renchi Yang. GraSP: Sim- ple yet Effective Graph Similarity Predictions. In AAAI,

    work page

  41. [2025]

    living room

    3 9 Supplementary materials Figure 5. Floor plan datasets and representation. Data Datasets: RPLAN and MSD We use RPLAN [35] and MSD [32]. RPLAN contains 88K+ floor plans and covers Asian residential apartments. As shown by [31], RPLAN contains a substantial amount of near-duplicates as well as floor plans that are not entirely connected ( i.e., floor pla...

    work page

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.