acceptodds
Under review as a conference paper at ICLR 2027

CityGRASP: Agent Driven Geo-Reasoning with Adaptive Spatial Programming for Open-Vocabulary Urban 3D Grounding

Abstract

Open-vocabulary city-scale 3D grounding localizes language-specified objects and regions in large urban scenes beyond predefined categories. However, the appropriate search scope and geometric computation often become clear only as contextual references and local geometry are resolved, while premature filtering can exclude the target. We formulate city-scale grounding as adaptive spatial programming, in which spatial search and geometric computation evolve jointly with execution evidence and uncertain decisions remain recoverable. We present CityGRASP, an agent-driven geo-reasoning framework that instantiates this formulation. Starting from query-guided open-vocabulary entity memory, progressive scene–program co-planning adapts the search scope and subproblem order as contextual references are resolved. Scene-conditioned geo-code synthesis composes low-level point-cloud and geographic primitives into computations tailored to local geometry, returning execution evidence to guide subsequent planning. To correct premature exclusions, reversible grounding certification checks query requirements and competing hypotheses, directs evidence revision, and tracks dependencies to restore affected alternatives while reusing valid preceding computations. Experiments on CityRefer, CityAnchor, and CitySTAR-3D demonstrate effective target disambiguation amid visually similar objects and complex spatial relations, with absolute [email protected] gains of up to 17.73% over the state of the art.

Primary area:
applications to computer vision, audio, language, and other modalities
Keywords:
Open-Vocabulary 3D Grounding, City-Scale Point Clouds, Agentic Reasoning, Progressive Program Synthesis

Reviews

Among:
Shape:
1234this paper, Top 15%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 3.00 over 3 reviews, above 85% of the others.
Reviewer_q1SmOpenReview ↗
Rating
3
Confidence
2
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

CityGRASP is an agent framework for open-vocabulary urban 3D grounding that dynamically adapts search and geometric code generation through execution feedback, using verification, revision, and selective recovery to correct earlier decisions and reuse valid computations. Experiments on three benchmarks report higher grounding accuracy than the evaluated baselines, while component ablations and restart comparisons support the effectiveness of its main designs.

Critical strengths

S1. The integration of search, geometric computation, and error correction is well motivated. The system uses execution feedback to adjust where to search, how to perform geometric computations, and whether earlier decisions need revision, while reusing computations that remain valid. This design fits the challenges of uncertain reference objects and complex spatial relationships in urban grounding. S2. The experiments provide substantial evidence that the components are useful. With the backbone models and initial candidates held fixed, removing progressive program generation reduces accuracy by approximately 28 and 34 percentage points in the two CityRefer settings. Selective recovery also reduces average runtime from 43.5 to 24.1 seconds compared with a full restart. These results support the practical value of the proposed designs.

Critical weaknesses

W1. The additional benefit of the specialized workflow over a general agent has not been directly established. The ablations mainly compare CityGRASP with reduced variants of its own system. They do not include a general agent that uses the same models and tools but autonomously organizes its planning and retries. Thus, the effectiveness of individual components is supported, but how much the complete specialized workflow adds over a general agent remains unclear. W2. Some key execution and comparison details are missing. These include how the system terminates and produces a prediction when verification remains unsuccessful, as well as the exact model and image processing procedure used for CitySTAR’s visual verification. These details affect the reproducibility of accuracy, computational cost, and baseline comparisons. This is currently a documentation and verifiability gap; there is no evidence that the reported results are incorrect.

Questions for authors

Q1. How much does the specialized workflow improve over a general agent under comparable conditions? Could you compare accuracy and actual computational cost while holding the language and vision models, basic tools, initial information, and budget limits fixed, and allowing a general agent to autonomously organize search, code execution, and retries? Q2. Could you clarify the visual model used in your CitySTAR baseline and provide a comparison with the same visual backbone? In your CitySTAR implementation, Qwen3-Reranker-4B is listed as the visual verification model, although Qwen’s official documentation identifies it as a text reranking model. Please clarify the exact checkpoint and how images are processed in your implementation. Since CitySTAR and CityGRASP already share Qwen3.5-9B and SAM3-based candidate construction, why were their visual backbones not aligned? Could you provide an additional comparison using the same visual model, such as Qwen3-VL-4B-Instruct, while retaining each system’s reasoning workflow? This would help distinguish improvements from the proposed workflow from those due to differences in visual model capabilities. Q3. When does the system stop, and what happens if verification remains unsuccessful? What are the inference and retry limits for each query? If the acceptance criteria remain unmet, how is the final prediction produced? What proportion of queries fall into this category, and what is their accuracy?

5 comments
AuthorsOpenReview ↗

Response to Reviewer q1Sm (1/3): Controlled Comparison with a General Agent

1. Controlled Comparison with a General Agent

We thank the reviewer for this suggestion and agree that a general-agent comparison is important for assessing the contribution of our specialized workflow. We have therefore added a comparison between CityGRASP and a general coding agent, reporting both grounding accuracy and actual computational cost.

The baseline follows a CodeAct-style interaction loop and uses the same Qwen3.5-9B reasoning model and Qwen3-VL-4B-Instruct visual model as CityGRASP. It invokes the same vlm_verify primitive and accesses the same entities, bounding boxes, maps and projected images. The agent autonomously organizes search, executes Python code, and retries or revises its actions using execution feedback, without following CityGRASP’s state machine.

In our initial runs, unrestricted exploration led to excessive runtime and occasional non-termination, while the agent made limited use of visual evidence. We therefore introduced landmark-first search with local inspection restricted to the 20 nearest candidates around the selected anchor, and provided projected candidate observations for appearance-based reasoning. We also added intermediate prediction saving, enforced termination under execution limits, and bounded request retries. Within these constraints, the agent determines its own sequence of searches, code executions and revisions. The general agent is limited to 16 outer model requests per query.

We evaluate both methods on 1081 CityRefer validation queries, comprising 579 NO and 502 ND queries, using the same 3D IoU protocol. We also report average runtime, model requests and token consumption per query.

MethodNO [email protected] (%)NO [email protected] (%)ND [email protected] (%)ND [email protected] (%)Time/query (s)Model calls/queryTokens/query (k)
General coding agent19.8619.1716.9316.73121.514.764.0
CityGRASP75.8274.6186.0684.2624.115.363.0

CityGRASP improves [email protected] by approximately 55.4 percentage points on NO and 67.5 percentage points on ND. It reduces average runtime by 80.2% with similar model-call counts and 1.6% lower token consumption. These results demonstrate a substantial advantage over the general coding agent and support the value of task-specific organization of agent-based coding and spatial reasoning for city-scale 3D grounding.

AuthorsOpenReview ↗

Response to Reviewer q1Sm (2/3): Visual Model Clarification and Matched-Backbone Comparisons

2. Visual Model Clarification and Matched-Backbone Comparisons

We thank the reviewer for carefully identifying this error. After checking the model-loading code and inference calls, we confirm that our CitySTAR baseline uses Qwen3-VL-Reranker-8B, which we mistakenly reported as the text-only Qwen3-Reranker-4B. We sincerely apologize for the confusion and will correct the model name throughout the manuscript.

We would also like to explain why we originally selected different visual models. For CitySTAR, we use the 8B reranker to score candidate-centered renderings against the language query, then combine these relevance scores with topology scores for final selection. For CityGRASP, we deliberately selected Qwen3-VL-4B-Instruct to evaluate intermediate visual constraints during agent-based coding and spatial reasoning. Its instruction-following capability also supports our broader goal of extending the agent to urban reasoning tasks beyond grounding.

Nevertheless, we understand and share the reviewer’s concern about backbone differences. We have therefore added CitySTAR–4B and CityGRASP–8B experiments to compare both methods under each shared visual backbone. We retain their respective reasoning workflows and keep Qwen3.5-9B, SAM3-based candidate construction, scene inputs, rendering, and preprocessing fixed.

For CitySTAR–4B, we normalize the Yes/No answer-token logits into a visual relevance score and retain the original topology fusion. For CityGRASP–8B, we adapt vlm_verify to calibrate the reranker’s per-view relevance scores using backbone-specific parameters. We then apply the original multi-view aggregation to obtain mean support and empirical bounds.

MethodVisual modelCityRefer NO [email protected] (%)CityRefer ND [email protected] (%)Time/query (s)
CitySTARQwen3-VL-Reranker-8B71.6880.6811.6
CityGRASPQwen3-VL-Reranker-8B75.1385.4627.9
CitySTARQwen3-VL-4B-Instruct66.8471.319.69
CityGRASPQwen3-VL-4B-Instruct74.6184.2624.1

CityGRASP retains its accuracy advantage on both splits under either shared backbone. Replacing CitySTAR’s original 8B model with 4B yields net losses of 28 NO and 47 ND correct predictions, reducing pooled accuracy by 6.94 percentage points. In comparison, replacing CityGRASP’s original 4B model with 8B yields net gains of 3 NO and 6 ND correct predictions, increasing pooled accuracy by 0.83 percentage points. These results show that CityGRASP is less sensitive to this visual-model substitution.

The runtime results also support our original model choices. Replacing CitySTAR’s original 8B reranker with the 4B instruction model reduces mean runtime from 11.6 to 9.69 seconds (16.5%), alongside a substantial accuracy loss. Replacing CityGRASP’s original 4B instruction model with the 8B reranker increases runtime from 24.1 to 27.9 seconds (15.8%), with a comparatively modest accuracy gain. CitySTAR therefore benefits substantially from the 8B reranker, while the 4B instruction model provides CityGRASP with a favorable balance of accuracy, efficiency, and flexibility.

AuthorsOpenReview ↗

Response to Reviewer q1Sm (3/3): Stopping Criteria and Unverified Predictions

3. Stopping Criteria and Unverified Predictions

We thank the reviewer for raising this important question. We understand the concern about how the system concludes when verification remains unsuccessful, and we agree that the stopping conditions and output rules should be stated explicitly.

CityGRASP accepts a prediction when the criteria in Section 3.5 are satisfied. These criteria require support for all necessary constraints under consistent reference bindings, every represented search branch to be explored or closed by evidence, and the applicable separation condition in Eqs. (4)–(5). Otherwise, the agent gathers additional evidence, revises references or search scopes, or repairs the program while preserving valid computations.

We impose a per-query execution budget and allow up to five recovery attempts. Each logical model request permits at most two retries after the initial attempt. Once the five recovery attempts are exhausted, the agent can continue ordinary search within the remaining execution budget. We terminate inference when that budget is exhausted, no admissible action remains, or an unrecoverable error occurs.

When inference ends without successful verification, we return a Top-1 prediction marked as unverified. We retain the current provisional prediction if available; otherwise, we select the candidate with the highest recorded support. If no usable score exists, we attempt visual scoring within the remaining budget and use deterministic candidate-ID ordering as the final fallback. We record the verification status and stopping reason.

To quantify these outcomes, we examined all 1081 CityRefer validation queries. The table reports the proportion of non-accepted queries and their [email protected], calculated within each non-accepted subset.

SettingTotal queriesNon-accepted queriesNon-accepted proportion (%)Correct non-accepted predictionsNon-accepted [email protected] (%)
NO57912822.118969.53
ND5025811.554374.14
Overall1,08118617.2113270.97

We interpret the higher non-acceptance rate in NO as reflecting the greater difficulty of establishing target–context correspondences for unseen instances. NO places greater demands on reference matching and spatial disambiguation, whereas ND evaluates new descriptions. Nevertheless, our returned predictions correctly localize 69.53% of non-accepted NO queries and 74.14% of non-accepted ND queries. A prediction can therefore identify the correct target even when the available evidence does not satisfy every verification condition.

Reviewer_97anOpenReview ↗
Rating
3(+1)
Confidence
2
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

The paper presents a framework for city-scale urban 3d grounding. Input to the system is a query from a user to find some object in the scene. This query can have plenty of complex relations like "across the road from ...", "closest to" anchored at some landmark, and the proposed system outputs an axis-oriented bounding box of the identified object.

The paper compares to various state-of-the-art methods (CityRefer, CityAnchor, CitySTAR) and shows both qualitatively (Figures 3, 4, 5) and quantitatively (Table 1) that it is more precise/robust in identifying the desired object described in the input queries.

Critical strengths
  1. According to Table 1, which is hard or even impossible to verify, the proposed method is better in identifying queried objects in a scene. The claimed gains, especially on CitySTAR-3D, are significant and would argue for the superiority of the proposed system.

  2. The appendix/supplementary material presents a lot of additional details, which I could however not completely verify because of time limitations.

Critical weaknesses
  1. The biggest weakness of the paper is that it is impossible to comprehend in my opinion. The whole Section 3 is very abstract and does not provide a guiding example or something which would give the user a chance to understand the proposed concepts. From this it is very hard to conclude whether the proposed methodology really makes sense.

  2. The qualitative comparisons are hard to decipher and interpret. In Figures 7 and 8 it is not even clear what the ground truth answer would be, and how far off the answers from the comparison methods are.

Some minor weaknesses: *) The paper has many typos, also in the figures. E.g., in Figure 1 it should say "Locate the flat-roof building", and not "Located the flat-root building". This reduces the perceived quality of the paper significantly, because what is worst is that this sentence full of typos was simply copied over without any checks to Figure 2 where it appears again.

*) "Comic Sans" is a font that in my opinion should not be used in a research paper and makes every figure look unprofessional and a bit childish.

*) Several mentionings of any percentage gains are incorrect, because the paper uses % when it actually means percentage points. E.g., in Table 1, the claimed gain from CitySTAR-3D to CityGRASP at [email protected] is 16.88 percentage points, and not %.

Questions for authors
  1. Just for reference: How would commercial models like Claude, ChatGPT, or Gemini perform in this task? I understand the value of using local models and would not demand the proposed system to perform better against the commercial models, but it would still be interesting to see.
5 comments
AuthorsOpenReview ↗

Response to Reviewer 97an (1/3): Method Clarification and Execution Example

1. Clarifying the Method and the Execution Example

We thank the reviewer for this comment and understand the reviewer’s concern regarding the need for a more concrete and intuitive presentation of the proposed method. We illustrate the proposed method through a concrete execution example in Figure 6, with accompanying explanations in Section 4.3 and Appendix A.4.2.

Briefly, CityGRASP locates a language-described target by progressively constructing and executing spatial reasoning programs. The agent determines where to search and writes geometric code to test the relevant spatial requirements. Execution results guide subsequent search and computation. If new evidence changes an earlier reference, the system revises the dependent computations and restores candidates excluded by that reference.

Figure 6 provides a complete execution example, accompanied by an explanatory paragraph in Section 4.3 (Lines 441–448) and further details in Appendix A.4.2. It shows how an incorrect identification of the Primark building causes candidate B to be excluded. Correcting this reference allows CityGRASP to restore B, recompute the affected steps, and obtain the final verified prediction.

We appreciate the reviewer’s suggestion and will revise Section 3 to make the explanation more concrete and accessible. We will explicitly connect the operations in Figure 6 to Sections 3.3–3.5, so that readers can follow the example alongside the method’s formal definitions.

AuthorsOpenReview ↗

Response to Reviewer 97an (2/3): Qualitative Results and Presentation Corrections

2. Improving Qualitative Figures and Presentation

We thank the reviewer for the detailed comments on the presentation.

Qualitative figures. We included Figures 7 and 8 at the end of the appendix to demonstrate CityGRASP’s effectiveness across a broader range of categories. For this purpose, we selected cases that CityGRASP successfully grounds. We have also provided more detailed comparisons, including ground-truth annotations and corresponding accuracy results, in Figures 3–5 of the main text.

Typos and typography. We will correct the wording in Figures 1 and 2 to “Locate the flat-roof building” and thoroughly check the manuscript, captions, and figure text for typographical errors. The current font was chosen to give agent interactions and execution workflows an illustrative style. We consider the font legible and consistent with these illustrations.

Percentage gains. We have thoroughly reviewed the manuscript and revised potentially ambiguous wording and notation to improve clarity for readers. Specifically, we now express absolute accuracy gains in percentage points, including the 16.88 percentage-point improvement on CitySTAR-3D at [email protected].

AuthorsOpenReview ↗

Response to Reviewer 97an (3/3): Commercial Models and General-Agent Comparison

3. Commercial Models and a General-Agent Reference

We thank the reviewer for this suggestion. Commercial models such as Claude, ChatGPT, and Gemini cannot directly process raw city-scale point clouds. Applying them to this grounding task requires substantial preprocessing and postprocessing to construct accessible scene representations and convert their outputs into 3D predictions. The resulting performance therefore reflects the combined capabilities of the model and the surrounding pipeline. Commercial APIs also incur additional usage costs, while service latency and backend execution conditions depend on the provider. We therefore did not include commercial models in this comparison and instead used locally deployed models. This allows us to keep the model configuration and execution environment consistent across methods and evaluate runtime and computational cost under controlled conditions.

We believe the reviewer’s underlying interest is in how much a specialized agent workflow improves over a general agent. We have therefore added a comparison with a general coding agent, which also addresses Reviewer q1Sm’s Q1.

The baseline follows a CodeAct-style interaction loop and uses the same Qwen3.5-9B reasoning model and Qwen3-VL-4B-Instruct visual model as CityGRASP. It invokes the same vlm_verify primitive and accesses the same entities, bounding boxes, maps and projected images. The agent autonomously organizes search, executes Python code and revises its actions using execution feedback. To support reliable execution, we introduced landmark-guided search over up to 20 nearby candidates, intermediate prediction saving and bounded request retries. The general agent is limited to 16 outer model requests per query.

We evaluate both methods on 1081 CityRefer validation queries, comprising 579 NO and 502 ND queries, using the same 3D IoU protocol. We also report recorded runtime, model requests and token consumption per query.

MethodNO [email protected] (%)NO [email protected] (%)ND [email protected] (%)ND [email protected] (%)Time/query (s)Model calls/queryTokens/query (k)
General coding agent19.8619.1716.9316.73121.514.764.0
CityGRASP75.8274.6186.0684.2624.115.363.0

CityGRASP improves [email protected] by approximately 55.4 percentage points on NO and 67.5 percentage points on ND. It reduces average runtime by 80.2% with similar model-call counts and 1.6% lower token consumption. These results demonstrate a substantial advantage over the implemented general agent and support the value of task-specific agent workflows for city-scale 3D grounding.

Reviewer_mmFQOpenReview ↗
Rating
3(+1)
Confidence
4
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

This paper proposes CityGRASP, an agent-based method for urban 3D grounding. The method sequentially plans the search scope and computation order, generates geometry computation code according to the scene, and restores candidates based on dependencies between evidence. Evaluation on three benchmarks, CityRefer, CityAnchor, and CitySTAR-3D, shows higher grounding accuracy than CitySTAR.

Critical strengths

CityGRASP achieves higher grounding accuracy than the recent method CitySTAR on multiple benchmarks.

Critical weaknesses
  • CitySTAR already provides an agent-based framework for urban 3D grounding. It is not sufficiently clear what new technical contributions the added sequential planning, geometry code generation, and candidate restoration provide beyond existing agent designs.
  • The ablation studies support the effectiveness of individual components. However, the visual models also differ in the comparison with CitySTAR, so the effect of this model change on the accuracy gap between the two methods remains unclear.
Questions for authors

For the sequential planning, geometry code generation, and candidate restoration presented as differences from CitySTAR, what is the technical novelty beyond incorporating standard replanning or backtracking?

2 comments
AuthorsOpenReview ↗

Response to Reviewer mmFQ (1/2): Methodological Novelty beyond CitySTAR

1. Comparison with CitySTAR and Clarification of Novelty

We thank the reviewer for highlighting the connection to CitySTAR. We recognize CitySTAR as an important framework for agent-based structured urban grounding. It introduces graph-structured CoT with tool-derived evidence, paired-hypergraph topology verification, and visual reranking.

CityGRASP represents a paradigm-level methodological advance over CitySTAR by formulating grounding as adaptive spatial programming. CitySTAR uses agents to coordinate existing tools for candidate acquisition and verification. CityGRASP extends the agent’s role to constructing and revising the spatial computations themselves, more fully exploiting the reasoning and coding capabilities of multimodal large language model agents. We realize this advance through three technical contributions:

  • Joint search–program construction (§3.3). We jointly determine the next search scope, reasoning subproblem, and required program inputs from execution evidence. The contribution is the coupling of spatial exploration with progressive construction of the reasoning program, extending CitySTAR’s adaptive orchestration of existing tools. Resolving a landmark can therefore change both where the agent searches and which geometric computation it constructs next.

  • Scene-conditioned geometric synthesis (§3.4). We enable the agent to construct the geometric implementation of a spatial relation, advancing beyond CitySTAR’s predefined relation checks. The generated code selects geometric inputs, local reference frames, and primitive compositions according to the scene. For example, across-road reasoning constructs a local road frame and evaluates opposite-side placement and road-footprint intersection. This allows the computation to adapt to local geometry while preserving the relation’s meaning.

  • Evidence-dependent recovery (§3.5). We make candidate exclusions and search-branch closures reversible by recording their supporting evidence. When a reference changes, these dependencies determine which decisions to revoke, which hypotheses to restore, and which computations to reuse. This enables recovery of targets excluded during earlier search, extending beyond CitySTAR’s verification of retained candidates. Reconstructing only the affected execution suffix also avoids repeating valid computations.

CityGRASP achieves 74.61%/84.26% NO/ND [email protected], compared with 71.68%/80.68% for CitySTAR. The controlled ablations support these contributions: replacing generated code with predefined relations reduces accuracy to 67.18%/76.89%, while removing scene conditioning yields 69.78%/79.68% (Table 2). Selective recovery reduces runtime from 43.5 to 24.1 seconds compared with Full Restart (Table 3). Our additional matched-backbone comparisons, reported in the second response, further confirm CityGRASP’s accuracy advantage under both shared visual models.

CityGRASP therefore advances beyond CitySTAR by enabling agents to construct and revise the spatial reasoning program itself. Its novelty lies in jointly adapting search and computation, synthesizing scene-specific geometric tests, and restoring excluded hypotheses through evidence dependencies while preserving valid computations. This constitutes a paradigm-level shift from orchestrating existing grounding operations to adaptive spatial programming, more fully exploiting the reasoning and coding capabilities of MLLM agents.

AuthorsOpenReview ↗

Response to Reviewer mmFQ (2/2): Visual Backbone Choice and Matched-Backbone Comparison

2. Visual Backbone Choice and Comparison

We thank the reviewer for this comment. We selected the visual backbones according to their roles in the two frameworks. In CitySTAR, we use Qwen3-VL-Reranker-8B to score query–candidate consistency for final visual reranking. In CityGRASP, the agent requests visual evidence during program construction and spatial reasoning. We therefore use Qwen3-VL-4B-Instruct to respond to these intermediate visual queries.

Our choice of an instruction-following model is deliberate: we aim to extend the agent to urban spatial reasoning tasks beyond grounding. Its general-purpose multimodal interface supports this broader objective, while the reranker is specialized for relevance scoring. We also observed lower visual-stage runtime with the 4B instruction model, making it suitable for repeated visual observations.

Nevertheless, we understand and share the reviewer’s concern about the different visual backbones. We have therefore added two experiments: CitySTAR with Qwen3-VL-4B-Instruct and CityGRASP with Qwen3-VL-Reranker-8B. We retain each method’s reasoning workflow and keep Qwen3.5-9B, SAM3-based candidate construction, scene inputs, rendering, and preprocessing fixed.

MethodVisual modelCityRefer NO [email protected] (%)CityRefer ND [email protected] (%)Time/query (s)
CitySTARQwen3-VL-Reranker-8B71.6880.6811.6
CityGRASPQwen3-VL-Reranker-8B75.1385.4627.9
CitySTARQwen3-VL-4B-Instruct66.8471.319.69
CityGRASPQwen3-VL-4B-Instruct74.6184.2624.1

CityGRASP outperforms CitySTAR under both shared visual backbones. With the 8B reranker, its NO/ND [email protected] gains are 3.45/4.78 percentage points. With the 4B instruction model, these gains are 7.77/12.95 percentage points. These comparisons support the contribution of the proposed reasoning workflow.

The results also support our original model choices. Replacing CitySTAR’s original 8B reranker with 4B reduces pooled accuracy by 6.94 percentage points, despite lowering runtime by 16.5%. In contrast, replacing CityGRASP’s original 4B instruction model with 8B increases pooled accuracy by only 0.83 percentage points, while increasing runtime by 15.8%. We therefore retain the 8B reranker for CitySTAR’s candidate scoring and the 4B instruction model for CityGRASP’s balance of accuracy, efficiency, and flexibility.

Comments on the paper

2 comments
AuthorsOpenReview ↗

Follow-up Before the Discussion Closes

Thanks again for your insightful comments and valuable time devoted to our paper. As the author-reviewer discussion period is coming to an end, please let us know if there's any further information we can provide to facilitate the discussion process. We are happy to answer any questions or concerns you may have during the author-reviewer discussion period.

Area_Chair_PGzZOpenReview ↗

Request for a brief response to the author rebuttal

Dear Reviewer mmFQ and 97an,

The authors have posted a detailed rebuttal addressing the concerns in your reviews. Reviewer q1Sm has already responded; I have not yet seen a follow-up from you.

I would be grateful if you could briefly indicate whether the rebuttal has addressed your main concerns, and whether you wish to update your assessment, before the deadline. Thank you very much for your time and for the reviews.

AC

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.