acceptodds
Under review as a conference paper at ICLR 2027

RoboEdit: Reference-Guided Local Video Editing for Robotic Scenes

Abstract

Robotic manipulation videos are costly to collect and often tied to a specific embodiment. This embodiment-specific coupling limits reuse across robot platforms. We study reference-guided local video editing as a way to repurpose such demonstrations by replacing manipulators or objects while preserving the surrounding context. Since paired cross-embodiment videos are rarely available, we construct pseudo-paired supervision by extracting masks and reference images from the source video itself. However, this construction introduces a train-test mismatch: at training time, the reference is a crop of the source manipulator and is therefore tightly coupled to the foreground, whereas at inference, the reference is a novel image of a different manipulator. This mismatch can lead to reference leakage, boundary artifacts, and original-identity ghosting. We propose RoboEdit, a mask-aware consistency learning framework that localizes reference-induced changes to the editable region. RoboEdit compares reference-conditioned and null-reference forward passes and suppresses reference-induced changes in outside-mask token relations. We derive this objective from a flow-matching view of vector-field decoupling and implement it efficiently with token-level reservoir sampling. To evaluate this setting, we curate a robotic local editing benchmark and introduce Robot Manipulation Consistency (RMC), a task-specific metric for measuring local stability, trajectory preservation, and edited-region semantic consistency. Experiments show that RoboEdit improves reference fidelity, scene preservation, and manipulation consistency over existing video editing baselines.

Primary area:
generative models
Keywords:
robotic video editing, reference-guided video editing, cross-embodiment transfer

Reviews

Among:
Shape:
1234this paper, Top 30%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 2.67 over 3 reviews, above 70% of the others.
Reviewer_QLCEOpenReview ↗
Rating
3
Confidence
2
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

RoboEdit proposes a method for ref-guided editing of robotic manipulation videos. This is to replace a robot's appearance while preserving the surrounding scene and motion. It fine-tunes HunyuanCustom using self-reconstruction examples built from existing videos, to avoid paired recordings across robot embodiments. Its core contribution is a token-correlation loss that compares ref-conditioned and null-ref passes and discourages changes in rel between tokens inside and outside the mask. Token sampling reduces the correlation computation. Recon and temporal losses provide extra supervision.

The paper evaluates the method on 335 video-ref pairs using visual metrics, a new robot manipulation consistency metric, human pref, and ablations. It also reports improved downstream control after pretraining visual representations on the output videos (average sim success reaches 41.1% compared to 33.2% for the strongest reported baseline). Some OpenArm experiments show gains too.

Critical strengths
  • The edited videos improve downstream control in simulation. Across 9 tasks and 3 robots, it achieves a higher average success than the strongest baseline that is reported. So seems like there is good evidence that the output videos have practical value for rep learning.
  • Ablation study is good and supports the claims... Adding the token-correlation loss improved DINO similarity and the scene-preservation judge scores. The remaining training setup is unchanged and so the comparison supports that the proposed loss is meaningful.
  • It addresses a data bottleneck and the setup is useful practically speaking. It builds supervision from existing videos using extracted masks and reference crops. So it avoids collecting matching demonstrations that repeat the same scene and trajectory with different robots (actual paired data).
  • Specifying the target robot through an image seems convenient. The method accepts a ref image and an edit mask without requiring a separate translator for every source-target robot pair or a calibrated simulator model of the target robot. This makes the setup attractive when suitable erndering assets or target domain recordings are unavailable. Generalization ot unseen robot identities still needs a clearer eval.
Critical weaknesses
  • It is not fully supported that the model can generalize to other unseen embodiments. The benchmark is described as OOD, but its separation from training data is not specified by dataset, trajectory, scene, task, or robot identity. It is especially unclear which target reference robot were absent during finetuning. This happens to be a central claim about handling new embodiments and is not fully supported by the experiments.
  • RMC doesn't directly establish that maniuplation remains correct. The metric combines mask overlap, robot-mask centroid agreement, temporal feature similarity, and ref matching. These can miss imperfect contact, object motion, or end-effector trajectories (or at least it is unclear how it won't miss). This matters because the manipulated object lies inside the regenerated region.
  • The main quantitative results are not extensive. Table 1 includes only HunyuanCustom, and it is unclear whether the baseline received comparable robot data finetuning. UniVideo and Wan 2.2 replace are excluded because of high failure rates without reporting rates. The controlled ablation supports the regularizer, but the broader comparison against existing editing methods is not fully conclusive.
  • The submission is not accompanied by video results. This is a highly visual method and it is close to impossible to actually validate the claims without being able to look at visual results.
Questions for authors

Clarification on the weaknesses mentioned above.

3 comments

Response to W1: Generalization to Unseen Robot Embodiments

We thank the reviewer for pointing out the need to clarify our out-of-distribution evaluation setting.

First, our evaluation benchmark is disjoint from the finetuning data. RoboEdit is finetuned on 57,343 real-world manipulation videos from seven OXE datasets. The evaluation benchmark contains 335 video-reference pairs from four separate sources (Jaco, KUKA, miscellaneous web videos, and self-collected laboratory videos), with no overlapping training episodes or trajectories.

Second, the benchmark includes both seen and unseen target robot embodiments. While some references correspond to robot families present during finetuning (e.g., KUKA, UR, and FANUC), others represent unseen robot types, including RealMan, Elephant, and Kinova. RoboEdit performs these replacements without target-specific finetuning or paired cross-embodiment demonstrations.

Third, our downstream simulation experiments provide additional evidence of cross-domain transfer. RoboEdit is finetuned entirely on real-world videos, while the downstream experiments use simulated Panda robots with appearances and rendering conditions absent from the finetuning data.

We have clarified the dataset splits, reference identities, and generalization settings in the revised manuscript: Section 5.3 now describes the held-out evaluation sources and the seen vs. unseen target references, and Appendix E ("Detailed Experimental Settings") adds an Evaluation Split and Unseen References paragraph detailing the training/evaluation split and the seen vs. unseen reference robots.

Response to W2: Object Motion and Interaction Preservation

We thank the reviewer for raising this concern. We agree that RMC primarily measures visual consistency and may miss errors in object motion or physical contact.

To address this, we conduct an additional evaluation in simulation, where ground-truth object trajectories are available. For each frame, we use the recorded 3D object position and camera parameters to determine where the manipulated object should appear. We then detect its position in the edited video and compare it with the corresponding GT-rendered video. Frames with unreliable GT detections are excluded.

MetricRoboEdit w/o reg.RoboEdit
Median Object Position Error (px) ↓7.144.61
Object Detection Rate (%) ↑72.082.7

With token-correlation regularization, object position error decreases by 35.4%, while detection rate improves by 10.7 percentage points. Unlike RMC's robot-mask centroid metric, this evaluation directly measures whether the manipulated object remains at its expected visual position.

Together with our downstream control ablations, these results provide additional evidence that the regularizer improves manipulation-relevant motion preservation. Although physical contact and end-effector accuracy are not directly measured, the new evaluation addresses object-motion errors that RMC may overlook (Appendix V).

Response to W3: Additional Quantitative Comparisons and Regularizer Ablations

We thank the reviewer for this suggestion. We have expanded the quantitative evaluation with additional editing baselines and regularizer comparisons.

Additional editing baselines (Appendix T). We quantify three failure modes of UniVideo and Wan2.2-Replace: original-arm residuals, unsuccessful replacement, and motion mismatch. Only 3.3% of UniVideo edits and 6.7% of Wan2.2-Replace edits pass all three checks in the sampled evaluation, explaining their exclusion from the main quantitative comparison.

Additional regularizer comparisons (Appendices S and U). We compare our token-correlation regularizer against two alternatives under matched training settings:

  • Direct feature consistency: RoboEdit reduces non-edit changed area from 9.82% to 8.26% and boundary changed area from 19.73% to 17.89%.
  • Global Gram alignment: On 316 paired cases, RoboEdit improves background PSNR by 0.77 dB, with gains in 92.4% of cases. Background SSIM and LPIPS also improve.

Downstream regularizer ablations (Appendices M and R). Removing the regularizer reduces frozen-encoder BC success by 9.5–13.4 percentage points across three source embodiments. In end-to-end π0.5 fine-tuning, target-domain success decreases from 43.0% to 29.0% without the regularizer.

These additional experiments strengthen the quantitative comparison and demonstrate that our regularizer improves both local scene preservation and downstream robot learning.

Response to W4: Supplementary Videos

We thank the reviewer for the suggestion. We have provided separate video files in the supplementary materials. We hope the new evidence adequately addresses the concerns and warrants a reassessment of the rating.

Thanks for the detailed rebuttal and more experiments. The new results, especially the object-motion eval and regularizer comps, strengthen the paper and address most of my initial concerns. Some limitations remain, especially around generalization, but I don't see them as major blockers. Overall, I'm satisfied with the rebuttal and will maintain my weak accept rating.

We thank the reviewer for the positive feedback and for acknowledging the additional experiments. We appreciate your constructive suggestions, which have helped us strengthen the paper.

Reviewer_8FZjOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

This paper presents RoboEdit, a framework for reference-guided local video editing in robotic manipulation scenes. The method uses pseudo-paired supervision constructed from existing robot videos and fine-tunes HunyuanCustom with masked reconstruction, temporal consistency, and token-level correlation objectives. Its main regularizer aligns outside-mask token relations between reference-conditioned and null-reference forward passes to reduce reference leakage into the surrounding scene. The paper evaluates editing quality on a curated benchmark and reports downstream visual representation transfer results in simulation and preliminary real-robot experiments.

Critical strengths
  1. The paper addresses a meaningful task. Repurposing existing manipulation videos through local robot replacement could help supplement costly real-world robot data and improve its reuse across embodiments.
  2. The proposed framework is technically plausible, and the qualitative results and quantitative evaluations are encouraging. The downstream experiments also provide useful evidence that the edited videos can benefit visual representation learning for robot control.
  3. The paper is clearly written and easy to follow, with a well-organized presentation of the motivation, data construction pipeline, and method.
Critical weaknesses
  1. The methodological increment over existing dual-branch Gram alignment is unclear. The cited work by Yan et al. (Towards High-resolution and Disentangled Reference-based Sketch Colorization) already uses shared-weight branches, a stop-gradient anchor, and Gram consistency to address reference-induced structural changes. RoboEdit introduces masking and token sampling, but lacks a direct comparison with the prior alignment objective under matched training conditions to establish the value of these adaptations.
  2. Mask robustness and manual processing costs are insufficiently evaluated. Segmentation errors may leave the original robot partially visible or expose nearby objects and contact regions to unintended changes. Although the pipeline includes mask dilation, filtering, and optional interactive refinement, the paper does not quantify sensitivity to mask errors or the required human effort, leaving its practical scalability unclear.
  3. The evaluation does not fully establish the core regularizer's benefit for manipulation. RMC uses robot-mask centroids and visual similarity, which may not reflect accurate end-effector motion or contact. The downstream results support the utility of edited data, but a comparison without the token regularizer and a more direct assessment of interaction preservation are needed to connect these gains to the proposed mechanism.
Questions for authors

The animations embedded in the PDF are difficult to inspect in common PDF viewers. I suggest providing the video results as separate files in the supplementary archive so reviewers can assess temporal consistency and contact-boundary artifacts over complete sequences.

4 comments

Response to W1: Comparison with Prior Gram Alignment

We thank the reviewer for highlighting this work. RoboEdit differs from Yan et al. in two aspects. First, their work focuses on image colorization, whereas robotic videos contain substantially more spatiotemporal tokens, making full Gram computation prohibitively memory-intensive. Token sampling is essential for feasible training: with 11,520 video tokens, it reduces 133 million correlation entries to 46,080 (2,880× fewer). Second, RoboEdit introduces mask-aware cross-region alignment, constraining foreground–background correlations while allowing changes within the edited region.

We compare against a global Gram baseline based on Yan et al., using the same video backbone, training settings. Unlike global alignment, RoboEdit constrains only correlations between inside-mask queries and outside-mask tokens. Results on 316 paired cases are:

MetricGlobal GramRoboEdit
Background PSNR ↑24.072524.8384
Background SSIM ↑0.91800.9219
Background LPIPS ↓0.06600.0638

RoboEdit improves PSNR by 0.77 dB, with gains in 92.4% of cases, alongside improved SSIM and LPIPS. These results support the benefit of mask-aware alignment over global Gram consistency under matched conditions (Appendix U).

Response to W2: Mask Robustness and Manual Processing

We thank the reviewer for raising this concern. Our pipeline generates masks automatically using SAM3, with optional positive/negative point refinement. Importantly, all masks used in our evaluations were generated automatically by SAM3, without manual correction.

To assess mask quality, we finetune π0.5 on three Robomimic tasks (Lift, Can, Square). We create a Teal Panda target domain by changing the robot's appearance while preserving its kinematics and dynamics. Each task uses 100 original Panda demonstrations plus 100 RoboEdit-generated sequences. We compare automatic SAM3 masks against simulator GT masks, keeping all other settings identical.

Target-domain success is 43.0% with SAM3 masks versus 45.3% with GT masks; source-domain success is 56.7% versus 57.7%. This small gap suggests that automatic masks capture most of the benefit of accurate segmentation without manual correction in this setting (Appendix R).

Response to W3: Regularizer Benefit for Manipulation

We thank the reviewer for this suggestion. We compare RoboEdit with and without token-correlation regularization through downstream control and object-motion evaluations.

Frozen-encoder transfer. Under identical MAE pretraining and frozen-encoder BC protocols, removing the regularizer consistently reduces success across nine manipulation tasks:

Source embodimentw/o reg.RoboEdit
Sawyer30.6%40.1%
IIWA29.6%43.0%
UR5e30.7%40.3%

End-to-end VLA fine-tuning. We fine-tune π0.5 on Lift, Can, and Square under the Panda-to-Teal-Panda appearance shift. Removing the regularizer reduces target-domain success from 43.0% to 29.0% (−14.0 points), compared with 25.7% for source-only training (Appendices M and R).

Object motion preservation. We agree that RMC may miss errors in manipulated-object motion. We therefore use simulator-recorded object trajectories and known camera parameters to determine where each object should appear. We compare detected positions in edited videos against corresponding GT renderings, excluding frames with unreliable GT detections.

Metricw/o reg.RoboEdit
Median Object Position Error (px) ↓7.144.61
Object Detection Rate (%) ↑72.082.7

Position error uses detected frames; detection rate uses all GT-valid frames.

The regularizer reduces object position error by 35.4% and improves detection rate by 10.7 percentage points. Unlike RMC's robot-mask centroid metric, this directly evaluates the manipulated object's visual position.

Together, these results show that regularization improves downstream control and object-motion preservation, although physical contact and end-effector accuracy are not directly measured. Detailed results are provided in Appendices M, R, and V.

Response to Question: Supplementary Videos

We thank the reviewer for the suggestion. We have provided separate video files in the supplementary materials.

Finally, we hope these additional results address the concerns and would appreciate reconsideration of the rating.

Dear Reviewer,

We would like to kindly follow up on our latest response, which includes additional experiments addressing your remaining concerns. We would greatly appreciate it if you could review these updates when you have a chance.

Thank you again for your time and thoughtful feedback.

Thank you for the additional experiments, which address several concerns. My main reservation remains the limited novelty of the adaptations to existing Gram alignment. Explicit segmentation also introduces potential mask errors and runtime overhead. I therefore maintain my weak reject rating. Separately, could the approach be adapted to edit human manipulation videos into useful robot demonstrations? Given the scarcity of robot videos, this could be a valuable future extension.

Thank you for your thoughtful feedback. While our method is inspired by prior Gram alignment, our contribution lies in adapting this idea to reference-guided local editing through mask-aware cross-region token correlations. Unlike global Gram alignment, our formulation selectively preserves the relationships between the edited region and its surroundings while allowing the target appearance to change. We believe this distinction is important for robotic manipulation videos, as further supported by our additional ablations.

We acknowledge the limitations of explicit segmentation and will clarify them in the revised paper. We also agree that human-to-robot video editing is a valuable direction, especially with the growing availability of egocentric manipulation data. Several recent works have explored this area, and we see it as a promising direction for future research.

Thank you again for your constructive feedback!

Reviewer_fX7mOpenReview ↗
Rating
3(+1)
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper presents RoboEdit, a reference-guided framework for locally replacing manipulators or objects in robotic videos while preserving the surrounding scene and motion. It uses self-reconstruction supervision from existing videos, avoiding paired cross-embodiment targets, and introduces a token-correlation regularizer to reduce reference leakage outside the edit mask. Experiments show improved editing quality and downstream visual representation transfer in simulation and a preliminary real-robot evaluation.

Critical strengths
  1. The problem is well motivated. Reusing existing manipulation videos across robot embodiments could reduce the cost of collecting embodiment-specific data.
  2. The paper offers a clever reformulation: it turns embodiment transfer without target-domain training videos into reference-conditioned masked video modeling. Self-reconstruction supervision avoids the need to collect or synthesize paired cross-embodiment target videos.
  3. The qualitative comparisons demonstrate the method’s advantages in reference fidelity and preservation of the surrounding scene.
Critical weaknesses
  1. The downstream evidence is limited to frozen-encoder behavior cloning, with a small-scale real-robot evaluation. These experiments support improvements in visual representation transfer, but leave it unclear whether the benefits extend to large-scale, end-to-end VLA training. This limits the evidence for the approach’s broader utility in robot learning.
  2. The main contribution is a token-correlation regularizer built on an existing video model. Its advantages over simpler feature-consistency or background-preservation constraints remain insufficiently established.
  3. Quantitative comparisons (Table 1) focus mainly on HunyuanCustom, with several alternative editing methods excluded.
  4. The work essentially leverages video editing for robotic data synthesis and augmentation. A more discussion of related work such as CosmosTransfer, RoboTransfer, and RoboEngine, would help better establish its novelty and practical advantages.
Questions for authors
  1. More comprehensive quantitative comparisons with state-of-the-art methods and further analysis would help substantiate RoboEdit's claimed advantages.
  2. The advantages of the proposed token-correlation regularizer over simpler feature-consistency or background-preservation constraints remain unclear and warrant further investigation.
4 comments

Response to W1 and Q1: Downstream Utility Beyond Frozen-Encoder BC

We thank the reviewer for this suggestion. We fine-tune π0.5 from a LIBERO checkpoint on Lift, Can, and Square (*Robomimic), with zero-shot success is 0% on all tasks. We evaluate a controlled appearance shift to Teal Panda, which shares Panda's kinematics and dynamics (a customized Panda arm created by us; please see Figure 13 in the revised manuscript and the simulation videos in the supplementary material). Each task uses 100 original Panda demonstrations (source-only baseline), optionally augmented with 100 sequences from HyCustom, RoboEdit (w/o regularization), and GT simulator rendering. Augmented videos retain the original action labels. Each policy is evaluated on 100 closed-loop rollouts per task and domain.

Training dataLift (P/T)Can (P/T)Square (P/T)Mean (P/T)
Source only83/7750/017/050.0/25.7
+ HunyuanCustom80/5737/2320/345.7/27.7
+ RoboEdit w/o reg.73/5542/2617/644.0/29.0
+ RoboEdit90/8353/3327/1356.7/43.0
+ GT rendering90/9063/4733/2062.0/52.3

Success rates (%). P: Panda; T: Teal Panda.

RoboEdit improves target-domain success from 25.7% to 43.0% (+17.3 points), outperforming HyCustom (27.7%) and recovering 65% of the observed GT-rendered augmentation gain. Removing our regularizer reduces success to 29.0%, demonstrating its downstream contribution. Source-domain success also improves from 50.0% to 56.7%. These results extend our evaluation to VLA adaptation and closed-loop control under controlled appearance shifts (Section 5.4, Appendix R).

Response to W2 and Q2: Token-Correlation vs. Feature Consistency

We thank the reviewer for this insightful question. We replace LtokL_{\mathrm{tok}} with direct consistency between normalized final-layer features near the outer mask boundary, keeping the remaining training setup unchanged. All variants are evaluated on 284 videos.

Metricw/o Reg.Feature Cons.RoboEdit
Boundary Changed Area ↓19.89%19.73%17.89%
Non-edit Changed Area ↓10.76%9.82%8.26%
RMC Trajectory ↑0.75820.77670.7877
FVD-Robot ↓85.5985.4687.27
VL Background ↑10.009.949.93

RoboEdit reduces non-edit changed area by 15.9% and boundary changed area by 9.3% relative to feature consistency. Feature consistency also improves spatial preservation over no regularization, while FVD and VLM background scores do not favor RoboEdit.

Training diagnostics show that outside-mask feature discrepancy decreases by approximately 50% even without regularization, while foreground–background correlation drift increases by 34%. This suggests that pointwise feature convergence does not necessarily preserve cross-region relationships, motivating our correlation-based constraint. The results support its advantage over the tested feature-consistency alternative (Section 5.5, Appendix S).

Response to W3 and Q1: Additional Baseline Evaluation

We appreciate the reviewer's suggestion to broaden the quantitative evaluation. We analyze three failure modes for UniVideo and Wan2.2-Replace: original-arm residual (patch-level NCC), replacement failure (DINOv2 similarity), and motion mismatch (optical-flow alignment).

Failure metricUniVideoWan2.2-Replace
Original-arm residual ↓23.3%46.7%
Replacement failure ↓33.3%16.7%
Motion mismatch ↓76.7%50.0%
Edit success ↑3.3%6.7%

Both methods exhibit over 90% failure rates on this sampled evaluation. UniVideo primarily struggles with motion preservation, while Wan2.2 frequently retains the original arm or alters its motion. This sampled analysis explains their exclusion from Table 1 (Section 5.3, Appendix T).

Response to W4: Related Work and Novelty

We thank the reviewer for highlighting these relevant studies, which help clarify the positioning of our work. We have expanded Section 2 to discuss Cosmos-Transfer1, which uses multimodal spatial conditions for controllable generation; RoboTransfer, which focuses on geometry-consistent multi-view synthesis; and RoboEngine, which emphasizes background augmentation for visual generalization.

RoboEdit instead targets single-reference-guided manipulator or object replacement, preserving the original motion and scene without paired cross-embodiment videos or target-specific simulator assets. Beyond applying video editing to robotic augmentation, we address reference leakage arising from pseudo-paired supervision through token-correlation regularization. The W2/Q2 ablation further supports this design.

We hope these new results address the concerns raised and a reconsideration of the assessment.

Dear Reviewer,

We would like to kindly follow up on our latest response and additional experiments. We would greatly appreciate it if you could take another look at our clarifications when you have a chance.

Thank you again for your time and constructive feedback.

Thank you for the detailed rebuttal and additional experiments. The new VLA evaluation and feature-consistency ablation address several of my concerns and provide stronger evidence for the method's effectiveness. While some limitations remain regarding methodological novelty and cross-embodiment generalization, I acknowledge these improvements and have raised my rating.

We sincerely thank the reviewer for their thoughtful feedback and for acknowledging our additional experiments. We are glad that the new results have addressed several of the initial concerns. We also appreciate the remaining suggestions, which will help us further improve the paper. Thank you again for your constructive review and for reconsidering your rating.

Comments on the paper

3 comments

We sincerely thank all reviewers for their constructive feedback and valuable suggestions. We have carefully addressed the concerns and strengthened the manuscript with additional experiments, analyses, and clarifications.

The main updates include:

  1. Downstream VLA evaluation: We fine-tune π0.5 and evaluate closed-loop control under a controlled appearance shift, demonstrating improved target-domain success (25.7% → 43.0%).

  2. Regularizer comparisons: We compare our token-correlation regularizer against direct feature consistency and global Gram alignment, providing further evidence of its effectiveness.

  3. Manipulation consistency: We introduce a simulator-grounded object-motion evaluation, showing a 35.4% reduction in object-position error and improved object detection.

  4. Supplementary videos: To address the difficulty of inspecting embedded animations in the PDF, we have uploaded separate videos showing real-world robotic video editing, simulation-based editing, and side-by-side comparisons with baseline methods. These videos allow reviewers to examine editing quality, temporal consistency, motion preservation, and contact-region artifacts over complete sequences.

  5. Additional evaluations and clarifications: We analyze failure rates of alternative editing methods, clarify evaluation splits and unseen robot references, and expand the related-work discussion.

All corresponding results and discussions have been incorporated into the revised manuscript and appendices. Detailed responses to each concern are provided in our individual replies.

We hope these additional results and clarifications address the reviewers' concerns and warrant a reconsideration of their assessments.

Thank you again for your time and thoughtful feedback.

Dear reviewers,

As the discussion deadline approaches, we wanted to follow up on our rebuttal, where we addressed the concerns raised in your reviews. We would appreciate it if you could take a look at our responses and reconsider your scores in light of them. If any concerns remain, we would be happy to clarify.

Thank you for your time and feedback.

Dear reviewers,

Thanks for serving as a reviewer. As the discussion period comes to a close and the authors have submitted their rebuttals, I kindly ask you to take a moment to review them and provide any final comments.

If you have already updated your comments, please disregard this message.

Thank you once again for your dedication to the OpenReview process.

Best,

Area Chair

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

Related papers

Loading the map…

Discussion (0)

Sign in to comment.