PolyDiffuse: Polygonal Shape Reconstruction via Guided Set Diffusion Models

This paper presents PolyDiffuse, a novel structured reconstruction algorithm that transforms visual sensor data into polygonal shapes with Diffusion Models (DM), an emerging machinery amid exploding generative AI, while formulating reconstruction as a generation process conditioned on sensor data. The task of structured reconstruction poses two fundamental challenges to DM: 1) A structured geometry is a ``set'' (e.g., a set of polygons for a floorplan geometry), where a sample of $N$ elements has $N!$ different but equivalent representations, making the denoising highly ambiguous; and 2) A ``reconstruction'' task has a single solution, where an initial noise needs to be chosen carefully, while any initial noise works for a generation task. Our technical contribution is the introduction of a Guided Set Diffusion Model where 1) the forward diffusion process learns guidance networks to control noise injection so that one representation of a sample remains distinct from its other permutation variants, thus resolving denoising ambiguity; and 2) the reverse denoising process reconstructs polygonal shapes, initialized and directed by the guidance networks, as a conditional generation process subject to the sensor data. We have evaluated our approach for reconstructing two types of polygonal shapes: floorplan as a set of polygons and HD map for autonomous cars as a set of polylines. Through extensive experiments on standard benchmarks, we demonstrate that PolyDiffuse significantly advances the current state of the art and enables broader practical applications.

Paper

References (51)

Scroll for more · 38 remaining

Similar papers

Peer review

Reviewer uTWe5/10 · confidence 4/52023-06-25

Summary

The paper presents a diffusion-based approach for reconstructing polygonal shapes from floorplan images and auto-drive sensor images. The idea is to refine an initial rough reconstruction through iterative denoising steps which revert a forward process that diffuses regular reconstructions into random noises through per-element noise injection (though the notions of elements are not clearly given for different tasks). The challenge with learning such a denoising diffusion model, as identified by the paper, is the robustness to permutations of orders of elements that are introduced by linearization (though it's not convincing why a set-based transformer model would introduce such ordering problem inevitably). The paper proposes to address this challenge by learning a guided forward diffusion process, such that the diffusion paths for different permutations are encouraged to be separate. In test time, the rough initial reconstruction is first converted to corresponding target noise distributions by the learned guided diffusion networks, and then reverted back to an accurate reconstruction through iterative denoising. The approach is tested on two polygonal shape reconstruction tasks, including floorplans and autonomous driving maps, and shows improvements over the baselines that produce initial reconstructions. Ablation studies are conducted on diffusion guidance and choices of several hyper parameters.

Strengths

The paper proposes to improve polygonal reconstruction works that produce results in a single network pass by iterative denoising of the reconstruction. It is thus expected that the results can have better quality than the initial estimates, which is confirmed by tests on two tasks. The paper formulates the diffusion model for structured elements like polygons and polylines, and identifies the problem of ordering permutation in representing the structures, which may cause severe ambiguity for a network to learn the reverse process. The paper designs a learned guided forward diffusion process to distinguish the different permutations causing representational ambiguity, and derives the updated diffusion models biased by the learned diffusions. Such adaptive and learned diffusion can be inspiring to other situations where hand-crafting a diffusion process can be problematic.

Weaknesses

The computational cost of the iterative process can be better illuminated, to help readers appreciate the comparative advantage obtained over the initial estimates produced by other methods. The paper identifies another challenge with diffusion-based approach for reconstruction: a reconstruction task has a single solution, so the initial noise needs to be chosen carefully. I do not fully understand this statement, and cannot find any validation of this challenge in the experiments and discussions. On the other hand, the proposed guided diffusion process does not produce a single noise for an input polygonal shape either, but samples from the predicted Gaussian distribution. The guided diffusion process is trained through contrastive learning on the initial reconstructions of training data by a different method. It is not clear how well the guided diffusion generalizes for data samples outside the training set. In contrast, a fixed diffusion procedure as adopted by DDPM and others does not have generalization issue, as the final noise distributions are known a prior. Indeed, the paper acknowledges that when the initial reconstructions come from hand annotations that are largely different from training reconstructions for HD Maps, the guided diffusion and denoiser do not work robustly. It might be difficult to regard this work as the first one using diffusion models for reconstruction. In fact, diffusion models have been widely used for conditioned generation that resemble reconstruction, e.g. ControlNet. Also, people have explored using diffusion models for segmentation [1], which is a kind of reconstruction. [1] SegDiff: Image Segmentation with Diffusion Probabilistic Models.

Questions

1. Computational costs of the iterative denoisers should be reported, and compared with the base cost of running the initial reconstructions. 2. Please provide clarifications and experimental evidence that perturbations of the optimal noise for a particular reconstruction can cause failure of denoise-based reconstruction. 3. In the footnote of page 3, it's stated that even a permutation-invariant model like Transformer has to pick a permutation as the representation of a data sample. This statement seems self-contradictory and needs clarification. For example, in DETR like works (including RoomFormer and MapTR) there is no ordering of the elements, and they rely on matching to compute the loss functions between two sets. Why cannot this approach to extended to the diffusion setting? 4. When defining the permutation contrastive loss (12), why not compare the x_t and x'_t, but compare x_0 and x_t instead? Is it essential to constrain the similarity between x_0 the uncorrupted data and x_t the noised version? This is not common for standard diffusion models. 5. Sec.4, feature representation, it's not clear how polylines/polygons of different vertex numbers are encoded and represented. This is an important detail as its encoding directly impacts how likelihood-based refinement (Fig4) can be done. 6. The line matching metrics augmented with angle threshold is not defined clearly. There is no formal description of its computation in either text or supplemental. 7. Define the standard DM model of Sec.5.3 clearly. Otherwise it's hard to justify any proposed component. 8. What is training schedule? It looks like training epochs but why would one bother with the training iterations so much. 9. Since it is acknowledged that the guided diffusion network may not generalize to hand-annotated input, I hope the authors can thoroughly discuss the limitations of the learned diffusion network, including how to decide if it's not general enough for a particular test dataset or input pattern, and what are the practical ways to enhance its generalization scope.

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

2 fair

Presentation

3 good

Contribution

2 fair

Limitations

Yes.

Reviewer Fcug6/10 · confidence 4/52023-07-02

Summary

This paper proposes a Guided Set Diffusion Model for reconstruction, addressing the challenges of ambiguous denoising and selecting appropriate initial noise. By learning guidance networks, the model ensures distinct representations for samples with multiple permutations in structured geometry. During testing, the model uses the guidance networks to initialize Gaussian noise and denoises to reconstruct polygonal shapes conditioned on sensor data. The approach is evaluated on floorplans and HD maps, demonstrating significant advancements over existing methods and enabling practical applications.

Strengths

The paper takes a step towards formulating reconstruction as a generation process conditioned on sensor data, providing insights into advancing Diffusion Models for shape reconstruction. The approach of guiding noise injection per element to ensure separation from permutation variants during the diffusion process effectively addresses the ambiguous denoising issue. The PolyDiffuse method is well-presented and achieves promising results for layout reconstruction. Overall, this paper makes valuable contributions and is commendable.

Weaknesses

While floorplan and HD map reconstruction are important problems, it appears that the GS-DM formulation is more suitable for layout or other semantic reconstruction tasks rather than real 3D shape reconstruction. It would be beneficial if the authors could provide a discussion on this aspect and given some suggestion of addressing the limitations of PolyDiffuse in extending to 3D shape reconstruction tasks.

Questions

N/A

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

N/A

Reviewer t57P6/10 · confidence 4/52023-07-05

Summary

This paper proposes a novel method for reconstructing multiple polygon shapes using a conditioned diffusion model. The method first learns a score-matching-based prior diffusion model from data. This model is then used to denoise the sensor data, resulting in the reconstruction of the polygon shapes. To handle the permutation order of the polygonal elements, the method separates the target Gaussian distribution of a sample from its permutation variants.

Strengths

1. A nice solution to handle the permutation order in the reconstruction of multiple geometric elements in conditioned diffusion model. 2. Comprehensive experiments to evaluate the performance of the proposed method. 3.This paper is well-written and easy to follow. Figure 2 provides a clear and concise illustration of the basic idea.

Weaknesses

1. The maximum number of polygons and their vertices used in the training of the diffusion model should be clarified. Are these numbers consistent with the ones listed in lines 183-184?

Questions

Overall, I like the idea of integrating a diffusion model as a prior into the reconstruction pipeline. However, I wonder if it is possible to handle the permutation order in a different way. For example, we could first train a network to predict the structure of the floorplan, such as the number of rooms and their basic bounding boxes. Then, we could denoise each shape individually. This approach might be easier to train and faster in inference.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

The authors have discussed the limitations of the proposed method in Sec. 7. I also want to know whether the denoising speed is fast enough for a real-time HD map reconstruction.

Reviewer pHyL7/10 · confidence 3/52023-07-06

Summary

- The paper introduces PolyDiffuse, a novel structured reconstruction algorithm that incorporates Guided Set Diffusion Models (GS-DM). - The core concept involves splitting the reconstruction pipeline into two distinct stages. - The forward diffusion process focuses on learning guidance networks to address denoising ambiguity effectively. - The guidance networks establish individual Gaussians as target distributions for each polygon and are learned prior to the denoising training phase. - The reverse denoising process utilizes the sensor data as a condition to reconstruct polygonal shapes. - The authors conduct extensive experiments involving two polygonal structure reconstruction tasks. - Through comprehensive quantitative and qualitative evaluations, the results demonstrate that GS-DM outperforms existing state-of-the-art methods in terms of performance and quality of the reconstructions.

Strengths

- The paper is well-written and easy to follow. - The derivation of the GS-DM is appropriately detailed. - By employing a multi-stage approach and introducing guidance, the proposed method effectively resolves ambiguity in polygonal shape scenarios. - The proposed method can evaluate reconstruction quality through likelihood evaluation. - The proposed method performs well even when using only rough annotations as input.

Weaknesses

The GS-DM may be sensitive to the specific guidance, as mentioned in Section 7 (Limitations).

Questions

- It is worth exploring whether incorporating the condition $y$ as an input to the guidance network leads to performance improvements. - What are the failure cases when applying PolyDiffuse to off-the-shelf methods? Is there further analysis regarding these cases?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

3 good

Contribution

3 good

Limitations

The authors addressed the limitations.

Reviewer HnVz7/10 · confidence 3/52023-07-07

Summary

The manuscript introduces a adaptation of the DDPM paradigm to enable denoising sets (instead of single data points like images). The method called GS-DM does this by introducing by adding noise per set element via learned guidance networks. This approach requires the addition of a proposal generator that initializes the number of sets and their parameters for the denoising process. The manuscript applies this new set-based diffusion model to two tasks: generating floorplan polygons from floorplan images and maps from autonomous car images. The proposed method performs well in comparison to related work.

Strengths

The authors identify a core limitation in DDPMs when it comes to set-based data and propose a sound way that empirically works well to solve it. The paper is well written, figures are high quality and support the writing well. The evaluation is thorough and clearly demonstrates that (1) the set-based diffusion is superior to the default DDPM (table 3) and (2) that the denoising approach helps improve state of the art results from MapTR and RoomFormer when applied on top of their final outputs. It is also very interesting to see how simple the inputs for denoising can be (as simple as the polygon centers) and still perform at almost state of the art levels. Since the underlying concept of set diffusion models is very general and could apply in various other domains, this work is significant to a larger audience than just the 3d reconstruction or perception community.

Weaknesses

While the paper is well written it could use some more clarity around how the guidance network is learned(line 122ff) since this is one of the key novelties. Just from reading the text I would not be able to reproduce the method at this point. - How is the neares negative permutation found? It seems like this could be an intractable problem very quickly? - is the triplet loss computed based simply on the polygon corners (for example?) The key limitation of the current approach is the fact that the number of elements in the set has to be initialized correctly by the proposal generator during inference. This is acknowledged by the authors and a reasonable limitation that can be addressed in follow on work imo.

Questions

- In Fig3: is it mu/sigma _phi or _psi?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

4 excellent

Contribution

4 excellent

Limitations

The limitations are adequately addressed by the authors at the end of the paper.

Reviewer uTWe2023-08-15

Thanks for the detailed response. I hope the authors can incorporate the updates and discussions to the final version.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC