acceptodds
Desk rejected at ICLR 2027

AsymmetricVAD: An Asymmetric Tri-modality Reasoning Paradigm for Weakly Supervised Video Anomaly Detection

Desk rejected at ICLR 2027, so left out of the paper lists and the map.

Abstract

Weakly supervised video anomaly detection (WSVAD) is a critical task in intelligent surveillance. It aims to detect and localize abnormal events using only video-level labels. However, traditional multimodal WSVAD methods predominantly rely on symmetric fusion paradigms that treat video, audio, and text modalities with equal importance. Such paradigms often overlook representation biases, making the joint features highly susceptible to volatile background audio noise and temporal alignment offsets. To address these challenges, we analyze the role of different modalities and propose AsymmetricVAD, an Asymmetric Tri-modality Reasoning paradigm with a conditional loop. Specifically, AsymmetricVAD utilizes a Selective Injection and Enhancement (SIE) module as an asymmetric gatekeeper. It selectively injects audio into the video/text feature space only under temporal fluctuation and high cross-modal similarity. Furthermore, a Look-Back mechanism establishes the loop to suppress the representation bias. To justify our method, we build a new dataset TMVAD, which is the first tri-modality dataset. Extensive experiments on the benchmark XD-Violence and TMVAD datasets demonstrate that AsymmetricVAD achieves competitive performance on coarse- and fine-grained WSVAD tasks. We will open-source our code and dataset at https://anonymous.4open.science/r/AsymmetricVAD.

Primary area:
applications to computer vision, audio, language, and other modalities
Keywords:
Video anomaly detection, Weakly supervised learning, Multimodal fusion, Cross-modal alignment

Reviews

Among:
Shape:
1234this paper, Top 49%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 2.33 over 3 reviews, above 51% of the others.
Reviewer_pM7zOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
3
Soundness
3
Read review
Paper summary

-- Key Idea:- The authors propose TMVAD, a new tri-modal video anomaly detection (VAD) dataset using video, text, and sound, containing 2000 videos across 4 coarse-grained and 22 fine-grained categories. Additionally, the paper highlights the role of audio in tri-modal Video Anomaly Detection and proposes a novel asymmetrical way to use the audio signal alongside the other modalities.

-- Results:- Experiments are conducted on two datasets, XD-Violence and TMVAD. The results outperform prior methods.

Critical strengths
  1. New trimodal dataset: The authors created a new video-audio-text dataset for Video Anomaly Detection, a domain where well-annotated data is very useful. The evaluation results tend to support the quality of the dataset, as the prior methods' results on TMVAD correlate with their performance on the reference dataset for the audio-visual domain, XD-Violence.

  2. Well-motivated approach: The authors examine the role of the audio modality in relation to text and video, highlighting the challenges associated with audio's sensitivity to noise, temporal misalignment with the image, and semantic divergence from the video. With this analysis, the authors justify the need for their asymmetrical method in tri-modal anomaly detection.

  3. Interesting and reliable method: The paper presents a new look-back strategy triggered by cross-modality discrepancy. This kind of safe strategy is welcome in a VAD model using audio.

  4. Generalizability via Plug-and-Play Integration: Beyond their main framework, the authors show in the appendix that both the SIE module and the Look-Back mechanism can be seamlessly integrated into existing baselines, providing consistent performance improvements. This highlights the general applicability and robustness of their approach.

Critical weaknesses
  1. Incremental Architectural Novelty: While the integration of an asymmetric pipeline for audio features and a Look-Back mechanism addresses cross-modal noise in VAD, the underlying technical components are not fundamentally novel. Asymmetric modality weighting and iterative refinement loops are widely used techniques in multimodal learning and vision architectures. In the current version of the paper, the contribution lies primarily in the application-specific assembly of existing briques and the creation of the TMVAD dataset, rather than a novel conceptual framework.

  2. Limited applicability in real-world surveillance & Over-reliance on the audio modality: In real-world VAD scenarios, such as standard CCTV networks, audio streams are rarely available due to hardware constraints and privacy regulations. The datasets used in this work (XD-Violence and TMVAD) heavily rely on web videos or movie clips containing rich audio tracks. Furthermore, while Table 3 demonstrates that audio improves the proposed model's performance, it also reveals a strong dependency: without audio (Config 1: V+T only, achieving 84.68% AP on XD-Violence), the architecture underperforms compared to existing V+T SOTA baselines reported in Table 1. The authors should discuss the practical limitations of this audio-centric paradigm and clarify how the model handles completely silent or audio-unavailable surveillance settings.

  3. Marginal Performance Gains on Coarse-Grained AP: The performance improvement on the XD-Violence benchmark for coarse-grained anomaly detection depicted in Table 1 is relatively narrow. By reporting the results of only one random seed, the authors make it difficult to determine whether this marginal gain represents a true statistical improvement or merely random variance during training. This issue is further compounded by a non-negligible computational cost, as the introduction of the SIE and Look-Back mechanisms increases the model's complexity to 14.03 GFLOPs, more than double that of competing frameworks like CMHKF (5.99 GFLOPs). If the performance gains of these components are neither statistically robust across random seeds nor efficient relative to their computational footprint, the practical value of adopting these modules in future SOTA models becomes questionable and weakens the paper’s claim of novelty.

  4. No information on synthetic audio generation for TMVAD: In Appendix B.1, “Tri-modality collection,” the authors mention that for videos lacking original audio in the TMVAD dataset, they generated synthetic audio and synchronized it with the video. The authors do not specify the method or model used for audio synthesis. An interesting analysis could be provided on how synthetic audio enhances or worsens detection compared to natural audio.

  5. Minor Formatting and Readability Issues in Tables 3 and 4: The presentation of the ablation studies in Tables 3 and 4 lacks consistency and creates unnecessary friction for the reader. In Table 3, the proposed full model (Config 4) is placed at the bottom, whereas in Table 4 (placed directly adjacent to it), the full model (Index 1) is placed at the top. Aligning the structural order of both tables would make comparisons much clearer. Furthermore, the “Config” and “Index” columns take up redundant space, as the remaining column descriptions are sufficient to identify each baseline in the text. Removing these redundant columns would improve visual clarity for the reader while saving layout space.

Questions for authors
  1. Could the authors highlight the specific theoretical or algorithmic innovations of the SIE and Look-Back modules beyond adapting these established principles to tri-modal VAD?

  2. Could the authors discuss the practical limitations of this audio-centric paradigm and clarify how the model handles completely silent or audio-unavailable surveillance settings? Also, often in video-audio surveillance settings, the audio part is very noisy and saturated. It is usually not used by security operators because it's not that informative.

  3. Could the authors report AP for XD-Violence and TMVAD on their method with 3 or 5 other random seeds to prove that these results are not due to variance?

  4. Could the authors provide information on the protocol they used to create synthetic audio streams for the silent videos?

  5. Given that synthetic audio is introduced for silent clips in TMVAD, could the authors clarify its influence on the paradigm? Ideally, applying this audio synthesis pipeline to a standard VAD benchmark like UCF-Crime would provide a valuable sanity check. We understand that conducting such experiments on UCF-Crime may not be feasible within the discussion period; in this case, could the authors provide a performance breakdown (both coarse- and fine-grained AP/mAP) separating videos with natural audio from those with synthetic audio in TMVAD? Additionally, it would be insightful to report whether the Look-Back mechanism is triggered at different rates on synthetic versus natural audio clips.

2 comments
AuthorsOpenReview ↗

We clarify the design motivation of our asymmetric paradigm and Look-Back mechanism, demonstrate robustness through multi-seed tests and qualitative demos, and provide a sub-group analysis of natural versus synthetic audio on TMVAD.

Response

[W1 + Q1]:

Thank you for giving us the chance to clarify.

While asymmetric weighting and iterative loops have been explored in broader multimodal learning, existing tri-modal VAD works (e.g., CMHKF) adopt symmetric fusion and one-way feedforward pipelines. We do not apply existing methods, such as uncertainty-based weighting in EAU (CVPR 2024) or recursive cross-attention in RJCMA (CVPRW 2024), because they still allow noisy acoustic feature channels to penetrate the joint latent space. To propose our asymmetric paradigm, we first conducted a systematic statistical analysis on surveillance datasets in main text’s Fig. 3, discovering that video-audio semantic discrepancy is over four times larger than video-text discrepancy, and over 78% of surveillance audio events exhibit temporal latency. Driven by these findings, we designed SIE, which formulates audio as a modulator to preserve visual-textual semantic orientation. Second, rather than relying on generic iterative refinement, we introduced Look-Back as a conditional closed-loop verification process that monitors global semantic disparity and selectively recalibrates biased feature channels only when significant conflict is detected.

Combined with our new TMVAD dataset, we believe our method provides valuable insights for future research.


[W2 + Q2]:

We acknowledge that many standard static CCTV setups may lack audio streams due to privacy or hardware constraints. However, our goal is to explore audio in scenarios where it is available.

As noted by the reviewer, our V+T configuration achieves 84.68% AP on XD-Violence. While it does not surpass the dedicated V+T methods, it remains competitive and outperforms methods like SteerVAD (ICLR2026) and HeadHunt-VAD (AAAI2026). Note that our paradigm is designed not to over-rely on audio. When audio is silent or unavailable, the model degrades to a V+T mode.

To handle noisy or saturated audio (as mentioned by the reviewer), our asymmetric design acts as a safeguard. The SIE module verifies cross-modal alignment and only uses audio as an auxiliary trigger. If the audio is pure noise, it fails the alignment check and is ignored.

To demonstrate this robustness, we have provided demo videos of our method compared to DSANet under two conditions (with audio and without audio) in https://anonymous.4open.science/r/AsymmetricVAD under the folder: demo-videos/ours_vs_DSANet.


[W3 + Q3]:

We evaluated our model over 5 random seeds.

Table 1: Multi-seed performance.

DatasetMetricCMHKF (ACL 2025)AsymmetricVAD (Ours)Margin ΔPaired t-test (p-value)
XD-ViolenceAP (%)86.60 ± 0.1687.24 ± 0.14+0.64p=0.003
XD-ViolencemAP@AVG (%)26.65 ± 0.2329.34 ± 0.18+2.69p<0.001
TMVADAP (%)86.91 ± 0.1888.20 ± 0.15+1.29p=0.001
TMVADmAP@AVG (%)27.18 ± 0.2428.32 ± 0.19+1.14p=0.002

Table 1 demonstrates that our training procedure is robust against random initialization.

Due to page limits, for our response regarding "Marginal Performance" in W3, please refer to our response to [W1] of Reviewer 1.


[W4 + Q4]:

For TMVAD, we use real audio from https://freesound.org/. Specifically, we select the audio corresponding to the actual abnormal event and then manually align it with visual events. Although we can not guarantee that it is the same as what the abnormal sound should be, we believe it can simulate surveillance scenes. We have attached demo videos in the folder: demo-videos/sample_TMVAD_synthetic.

We add the analysis in TMVAD Dataset section of the supplementary.


[W5]:

We have followed your suggestions in the revised manuscript.


[Q5]:

Compared to authentic audio, synthetic audio contains less accurate semantic representations of abnormal events, which slightly limits the joint cross-modal similarity calculation. However, as we discussed in the paper, we believe that leveraging synthetic audio is a practical approach to enhance the model’s robustness in surveillance scenarios, where audio may be absent in certain clips.

Following your suggestion, we provide the performance breakdown in Table 2.

Table 2: Sub-group analysis on TMVAD.

Audio SourceCoarse AP (%)Fine-grained mAP@AVG (%)Ratio (i=0)Ratio (i=3)
Natural Audio88.3828.4633.2%1.5%
Synthetic Audio87.8028.0430.4%1.2%
Full TMVAD88.2328.3532.1%1.4%

We find that AsymmetricVAD achieves consistent and comparable performance across both natural and synthetic audio. The performance drop in synthetic audio is because authentic audio provides more accurate semantic representations. Regarding Look-Back, the trigger rates on synthetic clips are lower than on natural clips because natural surveillance footage contains more complex background noise.

Dear Reviewer pM7z,

We truly appreciate your constructive comments helping us improve this paper.

We are sending a gentle reminder that the phase of the author-reviewer discussion is ending, and your problems mentioned in "Weaknesses" and "Questions" have already been addressed properly in both our first response and our revised paper. It would be much appreciated if you could kindly take some time to examine our response and tell us the remaining questions if any.

Thanks for your valuable time. May you have a good day.

Best wishes,
Authors

Reviewer_HtXUOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper proposes AsymmetricVAD, a weakly supervised video anomaly detection framework based on asymmetric video-audio-text fusion. It introduces Selective Injection and Enhancement (SIE) to identify informative audio signals and enhance visual-textual representations, and a Look-Back mechanism to recalibrate features when cross-modal semantic discrepancies are detected. The authors also construct TMVAD, a tri-modal dataset containing 2,000 videos across 22 anomaly categories. Experiments on XD-Violence and TMVAD report frame-level AP scores of 87.21% and 88.23%, respectively, along with improvements in fine-grained anomaly classification and temporal localization.

Critical strengths

1 The paper investigates audio noise and cross-modal misalignment in multimodal VAD, highlighting limitations of directly integrating heterogeneous features. The modality analysis provides useful empirical motivation for reliability-aware fusion. 2 The proposed framework reports performance gains on both XD-Violence and TMVAD across coarse-grained and fine-grained evaluation settings. The newly introduced TMVAD dataset also provides an additional benchmark for studying tri-modal anomaly detection.

Critical weaknesses

1 SIE and Look-Back primarily rely on established operations such as Top-K selection, cosine-similarity filtering, temporal smoothing, and feature gating. Their methodological distinctions from existing adaptive multimodal fusion approaches, particularly CMHKF, are not sufficiently established. 2 The proposed method achieves only marginal improvements over the strongest competing approaches. Moreover, the ablation studies indicate that SIE and Look-Back individually provide relatively modest performance gains, raising questions about whether their additional complexity is sufficiently justified.

Questions for authors

1 How does the SIE module fundamentally differ from existing adaptive audio extraction and fusion approaches, particularly the AFAE module in CMHKF? 2 In Eqs. (8)–(9), why is the original audio feature magnitude retained for segments outside the selected set? How does this design prevent irrelevant audio from affecting the visual and textual features? 3 Could the authors compare Look-Back with simpler single-pass feature-gating methods under matched settings and report the additional computational overhead introduced by iterative recalibration?

2 comments
AuthorsOpenReview ↗

We clarify the novelty of asymmetric design and the SIE module, justify our performance-complexity trade-offs, and demonstrate the effectiveness of the Look-Back mechanism via comparisons with two single-pass gating.

Response

[W1 + Q1]:

We thank the reviewer for this insightful question.

The SIE module fundamentally differs from existing approaches in two major aspects:

  • First, our approach operates under a fundamentally different Asymmetric paradigm, despite using basic operations like temporal windowing and Top-KK selection. In our paradigm, audio works as an auxiliary trigger and it is not explicitly included in the final anomaly representation. In contrast, existing approaches such as CMHKF fuse all three modalities to construct the final representation. Even though CMHKF allocates weights to different modalities, and the audio may have a small weight, the role of audio does not change. Thus, their performance may degrade in complex scenarios, especially in noisy environments due to unstable audio features.
  • Second, SIE explicitly addresses surveillance audio latency. Our statistical analysis on XD-Violence shows that 78% of videos have audio latency larger than 100 ms. To handle this, we leverage a Gaussian kernel to spread the temporal influence of audio features. In contrast, AFAE applies a discrete rectangular window with uniform weighting, which may introduce temporal artifacts due to abrupt boundary shifts.

To make the comparison clear, we have provided the demo videos in our anonymous project repository: https://anonymous.4open.science/r/AsymmetricVAD. You can view them in the following folder: demo-videos/ours_vs_CMHKF.


[W2]:

First, our performance gains are consistent with current top-tier progress. For example, MoME (CVPR 2026) improves by +0.02% AP over its second-best DEN-VAD on XD-Violence, and GS-MoE (ICCV 2025) achieves the same AP as the compared method TSA. Considering this context, our gains of +0.25% AP and +0.55% AP over the second-best LPGB and DSANet on two datasets represent a meaningful improvement.

Second, the complexity increment is also acceptable for real-world deployment. As shown in Table 13 of our paper, parameter growth is only 0.47M over CMHKF and inference latency is 11.4 ms vs. 8.6 ms. The 2.8 ms increment is within the real-time requirements for real-world surveillance deployment.


[Q2]:

First, discarding unselected segments makes the network focus on narrow Gaussian-enhanced peaks and causes severe feature scale distortion. It may lead to overfitting and degraded generalization on unseen scenarios. In contrast, retaining the audio magnitude Ea(t)E_a(t) preserves continuous background context while keeping the normal segments stable.

Second, because initial video and text representations may not be well-aligned, discarding the audio magnitude Ea(t)E_a(t) would destroy the raw audio features, which makes it impossible for the model to re-examine subtle anomaly features when updated video and text features are fed back into the SIE module in subsequent iterations.

To prevent irrelevant audio from affecting the visual and textual features, we generate a sharp contrast in modulation strength between anomaly events and normal background. Specifically, in Eq. (8), the Gaussian enhancement is strictly conditioned on both the audio saliency and the verified cross-modal alignment. This implies that only segments passing the Top-KK and similarity filtering receive the Gaussian enhancement. In Eq. (9), audio serves as an asymmetric scaling factor to modulate the video and text features. For irrelevant segments, the absence of the Gaussian enhancement leaves their low background energy Ea(t)E_a(t). In contrast, abnormal segments receive an enhancement that amplifies the corresponding video and text features. This relative disparity ensures that irrelevant audio noise is naturally marginalized.


[Q3]:

We replace Look-Back with two representative single-pass gating variants: REBA-style gating (CVPR 2026), which relies on a single-stage cross-attention gate, and RefineVAD-style gating (AAAI 2026), which employs a one-shot score-guided residual gate.

Table 1: Look-Back vs. Single-Pass Gating Methods on XD-Violence.

Gating SettingAP (%)Params (M)Latency (ms)
Single-Pass Gating (REBA-style)85.9112.729.5
Single-Pass Gating (RefineVAD-style)86.0313.189.8
Look-Back (Ours)87.2113.4311.4

As shown in Table 1, our method achieves superior performance with improvements of 1.30% AP over REBA-style and 1.18% AP over RefineVAD-style, while introducing negligible additional parameters (+0.71M over REBA-style Single-Pass Gating, +0.25M over RefineVAD-style Single-Pass Gating) and increasing average latency by merely 1.6 ms (11.4 ms vs. 9.8 ms). This minor overhead occurs because the Look-Back loop is triggered only when semantic misalignment occurs.

Dear Reviewer HtXU,

The author-reviewer discussion phase is ending. Please do not hesitate to let us know if any other questions or clarifications are required. We hope our responses have satisfactorily addressed your concerns, and we would be grateful for your favorable consideration. We truly appreciate your time and look forward to further discussions.

Thank you for your valuable time. May you have a good day.

Best wishes,
Authors

Reviewer_nDuSOpenReview ↗
Rating
3
Confidence
4
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

This paper proposes AsymmetricVAD, which selectively incorporates audio information into video-text representations and uses a Look-Back mechanism to reduce feature misalignment. The authors also introduce TMVAD, a tri-modal anomaly detection dataset, and demonstrate improvements on XD-Violence and TMVAD.

Critical strengths
  1. The asymmetric treatment of audio is well motivated and offers a practical solution to noisy multimodal fusion.
  2. The new TMVAD dataset and comprehensive ablations provide useful experimental support.
Critical weaknesses

The improvements over strong baselines are relatively modest, raising questions about the significance of the added complexity.

TMVAD includes synthetic audio for videos without original sound, which may introduce artificial correlations and affect the validity of multimodal evaluation.

Questions for authors
  1. How does performance change when evaluating only videos with authentic audio?
  2. Are the improvements consistent across multiple random seeds?
2 comments
AuthorsOpenReview ↗

We address all concerns by clarifying our performance-complexity trade-off, providing new evaluations and details on TMVAD's authentic-audio subset across multiple random seeds.

Response

[W1]:

Thank you for your kind response and recognition of our work.

First, our performance gains are consistent with current top-tier progress. For example, MoME (CVPR 2026) [1] improves by +0.02% AP over its second-best DEN-VAD [2] on XD-Violence, and GS-MoE (ICCV 2025) [3] achieves the same AP as the compared method TSA [4]. Considering this context, our gains of +0.25% AP and +0.55% AP over the second-best LPGB and DSANet on two datasets represent a meaningful improvement.

Second, the complexity increment is also acceptable for real-world deployment. As shown in Table 13 of our paper, parameter growth is only 0.47M over CMHKF and inference latency is 11.4 ms vs. 8.6 ms. The 2.8 ms increment is within the real-time requirements for real-world surveillance deployment.


[W2 + Q1]:

We have evaluated our method on the TMVAD authentic-audio subset, and the results are presented in Table 1.

Table 1: Performance comparison on the TMVAD dataset (Full vs. Authentic Only).

MethodModalityTMVAD (Full) AP (%)TMVAD (Authentic Only) AP (%)TMVAD (Full) mAP@AVG (%)TMVAD (Authentic Only) mAP@AVG (%)
CMHKFV+A+T86.9586.7427.1326.98
DSANetV+T87.6887.6127.6827.64
AsymmetricVAD (Ours)V+A+T88.2388.3828.3528.46

We find that our method achieves higher performance on the Authentic Only TMVAD subset than on the Full TMVAD dataset, with improvements of 0.15% and 0.11% in AP and mAP, respectively. This is because authentic audio naturally carries more accurate semantic representations of abnormal events than simulated sound, which facilitates the joint cross-modal similarity calculation and allows the SIE module to identify and amplify the verified anomaly segments. In contrast, the performance of other two methods both drop in terms of these two metrics on the Authentic Only subset. For CMHKF, it can attribute to the paradigm that utilizes the audio even when it is noisy in some videos. For DSANet, as it does not rely on the audio modality, the minor decrement is generated by the distribution of the dataset.

For TMVAD, we use real audio downloaded from the open-source platform https://freesound.org/. Specifically, we select the audio corresponding to the actual abnormal event in the video and then manually align it with visual events. Although we cannot guarantee that it is exactly the same as what the abnormal sound should be, we believe it can effectively simulate real-world surveillance scenes. Thus, the audio contains only minor artificial correlations. To make it clear, we have attached demo videos of the original TMVAD with synthetic audio in https://anonymous.4open.science/r/AsymmetricVAD. You can view them in the following folder: demo-videos/sample_TMVAD_synthetic.


[Q2]:

We have evaluated our model and the baseline over 5 independent random seeds (seeds: 42, 112, 2642, 192, and 2027) to assess consistency and reproducibility. The multi-run statistical results (Mean ± Std over 5 seeds) are summarized in Table 2 below.

Table 2: Multi-seed performance.

DatasetMetricCMHKF (ACL 2025)AsymmetricVAD (Ours)Margin Δ\DeltaPaired t-test (pp-value)
XD-ViolenceAP (%)86.60 ± 0.1687.24 ± 0.14+0.64p=0.003p = 0.003
mAP@AVG (%)26.65 ± 0.2329.34 ± 0.18+2.69p<0.001p < 0.001
TMVADAP (%)86.91 ± 0.1888.20 ± 0.15+1.29p=0.001p = 0.001
mAP@AVG (%)27.18 ± 0.2428.32 ± 0.19+1.14p=0.002p = 0.002

The standard deviations of AsymmetricVAD remain consistently small across all metrics, demonstrating that our training procedure is robust against random initialization. Furthermore, our method consistently outperforms CMHKF across all seeds, and paired t-tests confirm that the gains are statistically significant (p<0.01p < 0.01).


References

  • [1] Sun, Bo et al. “Joint Learning of General and Diverse Patterns with Mixture of Memory Experts for Weakly-Supervised Video Anomaly Detection.” Computer Vision and Pattern Recognition (2026).
  • [2] Zhang, Chen Jason et al. “Dynamic Erasing Network with Adaptive Temporal Modeling for Weakly Supervised Video Anomaly Detection.” IEEE Transactions on Neural Networks and Learning Systems 36 (2025): 16706-16720.
  • [3] D'Amicantonio, Giacomo et al. “Mixture of Experts Guided by Gaussian Splatters Matters: A New Approach to Weakly-Supervised Video Anomaly Detection.” 2025 IEEE/CVF International Conference on Computer Vision (ICCV) (2025): 10275-10285.
  • [4] Joo, Kevin H. et al. “CLIP-TSA: Clip-Assisted Temporal Self-Attention for Weakly-Supervised Video Anomaly Detection.” 2023 IEEE International Conference on Image Processing (ICIP) (2022): 3230-3234.

Dear Reviewer nDus,

We truly appreciate your constructive comments that helped us improve this paper.

We are writing to send a gentle reminder that the deadline for the author-reviewer discussion phase is rapidly approaching, and we have meticulously addressed your concerns in our response and our revised paper. It would be much appreciated if you could kindly spend a few minutes going through our response and tell us if there are any remaining questions.

Thanks for your valuable time. May you have a good day.

Best wishes,
Authors

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

Related papers

Discussion (0)

Sign in to comment.