Pith. sign in

REVIEW 4 major objections 6 minor 86 references

Implicit Counterfactual Learning for Audio-Visual Segmentation

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Introducing implicit text and diffusion-based counterfactuals into audio-visual segmentation training yields state-of-the-art masks on S4, M3, and AVSS.

desk verdict Bad sign error in the central CDCL loss makes the written method invert its own objective; the underlying idea and ablations are worth a serious look, but the empirical claims need a corrected loss or code. read the letter →

arxiv 2507.20740 v1 pith:ZAR2RQRA submitted 2025-07-28 cs.CV

Mingfeng Zha , Tianyu Li , Guoqing Wang , Peng Wang , Yangyang Wu , Yang Yang , Heng Tao Shen This is my paper
classification cs.CV
keywords audio-visualsegmentationimplicitcounterfactuallearningmulti-granularitytextlatentdiffusioncounterfactualsorthogonalrepresentationdistribution-awarecontrastivemodalityimbalancecausaldebiasing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that audio-visual segmentation fails not mainly from weak fusion but from two structural problems: visual and audio features live in incompatible spaces, and visual features dominate decision-making. To fix both, it proposes the implicit counterfactual framework (ICF), which retrieves multi-granularity text descriptions matched to video, segment, and frame features, uses a latent diffusion model to generate counterfactual texts orthogonal to the factual ones, and aligns all modalities through distribution-level contrastive learning. The claim is that this training-stage recipe removes spurious visual priors and corrects audio-visual mismatch, yielding state-of-the-art J&F of 90.07, 69.89, and 48.16 on the S4, M3, and AVSS benchmarks. A careful reader should care because the method is presented as a plug-in that raises other AVS models by 3–4% without changing their test-time inference.

What carries the argument

The argument is carried by three coupled mechanisms. Multi-granularity implicit text (MIT) retrieves learnable text embeddings matched at video, segment, and frame level for visual features (VideoCLIP/CLIP) and at audio level (CLAP), then fuses them through a gate into a composite factual text z that is supposed to live in a modality-shared semantic space. Semantic counterfactual (SC) adds noise to z, orthogonalizes part of the noise against z with Gram-Schmidt, and denoises with a latent diffusion model to produce counterfactual texts z*, with an orthogonality loss keeping them similar enough to be relevant and different enough to be counterfactual. Collaborative distribution-aware contrastive learning (CDCL) turns visual, audio, and text features into Gaussian distributions with mean and covariance, measures distance with Wasserstein distance plus entropy terms, and contrasts factual–counterfactual and inter-modality pairs. Together they are meant to correct visual bias and provide external semantic grounding.

What would settle it

Measure the agreement between the visual implicit text lv (retrieved by VideoCLIP/CLIP from frames) and the audio implicit text la (retrieved by CLAP from sound) on a labeled subset of M3 and AVSS where the sounding category is known; if the two retrievals do not name the same sounding object at well above chance, the shared text space grounding the counterfactual contrast does not exist. A complementary check is to replace the composite text z with random vectors of the same dimension and see whether J&F drops materially.

Watch

Extended reading notes

Core claim

ICF is a training-only framework that treats text as a shared semantic surface between vision and audio. For a given video, the model retrieves implicit text embeddings at three granularities from the visual side (via VideoCLIP and CLIP) and one from the audio side (via CLAP), fuses them into a composite factual text z, then pushes z partway through a latent diffusion process, orthogonalizes random directions in the intermediate latent space with Gram-Schmidt, and denoises to obtain counterfactual texts z*. A collaborative distribution-aware contrastive learning loss models visual, audio, and text features as Gaussian distributions and separates factual from counterfactual pairs while pulling matched modalities together. The paper reports that this scheme outperforms the previous best method by 1.77 points on S4, 4.69 on M3, and 4.06 on AVSS with a PVT-v2 backbone, and that inserting ICF into two existing AVS models improves their J&F by about 3–4 points.

Load-bearing premise

The load-bearing premise is that the implicit texts retrieved from vision and audio live in one shared semantic space, so the fused text z is a true bridge between modalities; the paper never verifies that VideoCLIP/CLIP and CLAP embeddings are aligned.

Editorial extensions

If this is right

  • On multi-source and semantic benchmarks (M3 and AVSS), the reported gains are larger than on single-source S4, implying the method helps most where audio-visual association is ambiguous.
  • ICF is applied only during training, so adding it to an existing AVS model changes the training objective without adding any test-time text construction or inference overhead.
  • Because the counterfactual texts are generated end-to-end through retrieval plus diffusion, the method avoids the storage and annotation costs of explicit LLM-generated captions, audio replacements, or pair swapping.
  • The distribution-level contrast operates on Gaussian statistics rather than individual feature vectors, which the paper argues makes alignment robust to blank frames, mixed scenes, and audio noise.
  • Plug-in experiments show the framework transfers to other architectures: AVSBench and AVSegFormer both improve without architectural change.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the shared-text-space assumption holds, a cheaper general recipe for debiasing audio-visual segmentation is implied: pretrained contrastive text encoders plus a small diffusion module, with no explicit captions, no manual attribute editing, and no extra inference cost.
  • Because the paper's ablation shows segment-level semantics contributes the largest gain, making the granularity weighting adaptive per video or temporal segment is a natural extension that could cut compute while preserving accuracy.
  • The paper's stated O(n^2) Gram-Schmidt cost and hand-tuned hyperparameters suggest replacing orthogonalization with SVD and learning the control parameters would be a testable scalability improvement.
  • The claim that implicit texts resist low-level noise could be checked directly by corrupting frames or adding interfering audio at test time; if ICF degrades gracefully, the distribution-level contrast is the active robustness mechanism.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes an implicit counterfactual framework (ICF) for audio-visual segmentation, consisting of three components: multi-granularity implicit text (MIT) as a semantic bridge, semantic counterfactual (SC) via diffusion and orthogonalization, and collaborative distribution-aware contrastive learning (CDCL). The authors report state-of-the-art results on AVS-Object (S4 and M3) and AVSS, and show that ICF can be plugged into other AVS models with consistent gains. The experimental section includes component ablations, hyperparameter sensitivity studies, qualitative comparisons, and t-SNE visualizations.

Significance. If the method as described were correct, the contribution would be significant: it is one of the first attempts to use implicit text and counterfactual generation for audio-visual segmentation, and the reported plug-in gains suggest a reusable training objective. The paper is well positioned in the literature and the empirical study is broad. However, the central CDCL loss as written has an inverted contrastive sign, and the Gram-Schmidt update in the SC module is dimensionally inconsistent. These are load-bearing issues because CDCL and SC are the main claimed technical novelties and the ablations attribute substantial gains to them. In addition, the visual-text contrast is potentially self-referential, and the absence of code and error bars makes the empirical claims difficult to verify. The manuscript therefore needs substantive revision before the results can be trusted.

major comments (4)
  1. [§3.4, Eq. (16)] The definition D′=exp(D(·,·)/τ) inverts the intended contrastive semantics. Since D is a distance, exp(D/τ) increases as the distance increases. The loss −1/B Σ_i log [ Σ_{j∈P(i)} D′(F_i^v,A_j^+) / (Σ_{j∈P(i)} D′(F_i^v,A_j^+) + Σ_{k∈N(i)} D′(F_i^v,A_k^-)) ] is minimized by increasing D′(F_i^v,A_j^+) relative to D′(F_i^v,A_k^-), which means the optimization increases the distance to positive pairs and decreases the distance to negative pairs. The same sign error propagates to Eq. (17) and to the analogous audio-text loss L_{a↔l}. Since CDCL is a central contribution and Table 2 attributes a large gain to it, the reported improvements are not explained by the written loss. If the implementation uses exp(−D/τ), then Eqs. (16)–(17) must be corrected; if it uses exp(D/τ), the loss would repel the very pairs it is meant to attract.
  2. [§3.3, Eq. (7)] The Gram-Schmidt update is dimensionally inconsistent. The right-hand side ∥r(i) − (r(i)·z(i))z(i)∥ is a scalar norm, while the left-hand side r(i) is a vector. As written, the equation cannot define an orthogonalized random vector or matrix. The intended operation is presumably to subtract the projection of r onto z and then normalize, but the paper does not provide a well-defined formula. This matters because SC is one of the three main modules, and its design is directly used in Eq. (8) and the counterfactual pool construction in Eq. (12).
  3. [§3.2–§3.4, Eqs. (3)–(5), (17)] The visual-text contrast is potentially self-referential. In Eq. (3), ℓv is retrieved as the argmax of similarity between Fv and VideoCLIP text embeddings, so ℓv is a function of Fv; Eq. (4) fuses these retrieved texts, Eq. (5) builds z from them, and Eq. (17) uses z as the positive anchor for Fv. Consequently, the positive pair ⟨Fv,z⟩ can be made close by a text embedding that is essentially a transformed version of the visual features themselves, without any independent semantic grounding. The paper does not verify that VideoCLIP/CLIP text embeddings and CLAP text embeddings lie in a shared space, nor that z corrects rather than inherits visual bias. The unbiasedness claim requires a concrete test, such as ablating ℓv with random or fixed text tokens or reporting a zero-shot text-anchored retrieval experiment.
  4. [§4, Tables 1–3 and Figure 5] The empirical verification is incomplete: no error bars or statistical significance tests are reported, and no code is released. Many ablation differences in Table 2 are around 0.3–1.0 J&F, which may be within run-to-run variation. Since the paper's main evidence is empirical and the central loss has a sign ambiguity, the authors should either release code or report mean and standard deviation over multiple seeds so that the reported gains and ablations are checkable.
minor comments (6)
  1. [§3.2, Eq. (5)] The fusion function F in Eq. (5) is not defined; please specify whether it is an MLP, a transformer layer, or another module.
  2. [§3.3, Eq. (8)] The per-sample parameters m_i and s_j are introduced in Eq. (8) but are never specified or ablated; later they are effectively disregarded when deriving the weights in Eq. (17), so the claimed intra-sample and inter-sample variation is not directly evaluated.
  3. [§4.1, Table 1] The use of external foundation models is marked with a checkmark for several methods, but the implications for fair comparison are not discussed; the proposed method uses multiple external models (VideoCLIP, CLAP, latent diffusion), so the cost and benefit relative to methods without such models should be clarified.
  4. [§4.2, Table 2] The row labels 'Feature-level* Inter-sample Intra-sample' and 'Discrete Continuous Lortho' are unclear; please define these dimensions explicitly so that the ablation structure is easy to follow.
  5. [Figure 5] Several axis labels in Figure 5 are truncated (e.g., 'Change of and ' should be 'Change of J and F'); please provide complete labels and preferably include confidence intervals or seed variation.
  6. [References] DiffusionAVS [46] is cited as an arXiv preprint; if a peer-reviewed version exists, please update the reference. The mention of TeSO in Section 4.1 uses a citation marker but is not included in Table 1; please make the comparison consistent.

Circularity Check

1 steps flagged · score 4.0 of 10

Vision-side implicit text is retrieved from the visual features it later supervises, making the MIT/CDCL text alignment partly self-referential; the benchmark results remain external.

  1. self definitional [Sec. 3.2, Eqs. (3)-(5); Sec. 3.4, Eq. (17)]
    "lv ∈ arg max_{l1,l2,...,lk} 1/kt Σ_{i=1}^{kt} sim(Fv, VideoCLIP(li)) (3) ... ℓv = Σ ... (4) ... z = F(Concat(Gate(ℓv), Gate(ℓa))) (5) ... Lv↔l = −1/B Σ_i log [ D′(Fv_i, zi) / (D′(Fv_i, zi) + Σ_k w_k D′(Fv_i, ℓcf_k)) ] (17)"

    The vision-side factual text ℓv is defined as the text embedding(s) maximizing similarity to the visual features Fv (Eq. 3), then fused into z via Eqs. (4)-(5), and z is used as the positive anchor in the visual-text contrastive loss (Eq. 17). Thus the positive pair ⟨Fv,z⟩ is close by construction: the anchor is selected from the very features it is supposed to align. Optimizing Lv↔l does not ground vision in an independent semantic space; it only reinforces the retrieval choice already made from Fv. The claimed 'semantic bridge' or 'prior guidance' of MIT is therefore partly a self-consistency loop rather than an external text prior. The audio side is analogous, since ℓa is matched from audio features via CLAP and then contrasted against audio features through z.

full rationale

The clearest circular step is the construction of the vision-side text anchor: ℓv is retrieved by maximizing similarity to Fv (Eq. 3), incorporated into z (Eq. 5), and then used as the positive anchor for Fv in CDCL (Eq. 17). This makes the visual-text alignment self-referential by construction. However, the paper's headline results are external benchmark comparisons on AVS-Object and AVSS, and the plug-in improvements in Table 3 are empirically measured against other models, so the central empirical claims are not forced by the self-referential anchor. The diffusion-based counterfactuals and CLAP-based audio text provide some independent signal. There is no load-bearing self-citation or imported uniqueness theorem. The alleged sign inconsistency in Eqs. (16)-(17) is a correctness concern, not a circularity concern, and is not scored here. Overall, the circularity is real but partial and localized to the claimed semantic-grounding role of MIT/CDCL, warranting a score of 4.

Assumptions & free parameters 8 free parameters · 6 assumptions · 2 invented entities

The central claim rests mainly on pretrained cross-modal encoders, a diffusion prior, and several manually tuned strengths and loss weights. No causal model is specified; the counterfactual samples are defined by orthogonal projections in a latent space.

free parameters (8)
  • alpha (counterfactual strength) = αo in [0.7, 0.8) (Fig. 5)
    Controls orthogonality ratio in Eq. 8; tuned per dataset.
  • m_i and s_j (per-sample variation) = not specified
    Inter-sample and intra-sample variation factors in Eq. 8; ad hoc.
  • lambda_z = not specified
    Balances similarity and orthogonality in Eq. 10.
  • loss weights lambda_cf, lambda_CDCL, lambda_p_to_q = not specified; deferred to supplementary
    Total loss weighting in Eq. 19; tuned on validation.
  • Top-K counterfactual pool size kc = 8
    Selected by ablation in Fig. 5.
  • diffusion steps sd = 200
    Selected by ablation in Fig. 5.
  • temperature tau = not specified
    Used in softmax correlations Eq. 1 and contrastive losses.
  • learnable implicit text tokens l and fusion weights w = learned during training
    Eq. 3-4; not hyperparameters but fitted parameters the method depends on.
assumptions (6)
  • ad hoc to paper Gram-Schmidt orthogonalization in Eq. 7 preserves enough semantic information that the denoised z* acts as a meaningful counterfactual.
    The method assumes orthogonalizing a portion of the latent factual text creates alternative semantic samples rather than destroying structure; no evidence for this semantic claim is given.
  • domain assumption CLIP, VideoCLIP, and CLAP embeddings form a common modality-shared space for vision, video, audio, and text.
    Required in Sec. 3.2 for ℓv and ℓa to be fuseable and for text to bridge audio and visual features; the paper does not quantify cross-encoder alignment.
  • domain assumption Gaussian modeling of frame-level features with Eq. 13 captures the semantic distribution of a clip.
    The paper replaces feature points with mean and covariance; this assumes temporal features are roughly Gaussian.
  • domain assumption Latent diffusion forward/reverse process from [53] and the noise prediction objective can be conditioned on visual features without audio-specific adaptation losing semantics.
    Relies on a pretrained LDM to denoise text latents; no fine-tuning details are given.
  • ad hoc to paper Counterfactual reasoning from Pearl's ladder applies to orthogonalized latent text even though no structural causal model or intervention is defined.
    The causal language in Sec. 1 and Sec. 3.3 is not backed by an explicit causal graph; this is a conceptual assumption.
  • standard math Standard mathematical background (Wasserstein distance, Gaussian entropy, diffusion formulation) is assumed correct.
    Background results from [53,55,57] are used without proof.
invented entities (2)
  • Multi-granularity implicit text (ℓv, ℓa)
    purpose: A latent text-like representation retrieved from video, segment, and frame features to bridge audio and visual modalities.
    No external semantic ground truth or independent test of text quality; its only validation is downstream segmentation accuracy.
  • Counterfactual text pool (z*)
    purpose: Negative samples for contrastive learning, generated by orthogonalizing z in the diffusion latent space.
    These are algorithmic constructs; the paper provides no falsifiable prediction outside the method itself.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Implicit Counterfactual Learning for Audio-Visual Segmentation." pith.science (2026). https://pith.science/paper/ZAR2RQRA

@misc{pith2026250720740,
  author       = {Pith},
  title        = {Pith review of: Implicit Counterfactual Learning for Audio-Visual Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZAR2RQRA}},
  note         = {Machine review of arXiv:2507.20740}
}
read the original abstract

Audio-visual segmentation (AVS) aims to segment objects in videos based on audio cues. Existing AVS methods are primarily designed to enhance interaction efficiency but pay limited attention to modality representation discrepancies and imbalances. To overcome this, we propose the implicit counterfactual framework (ICF) to achieve unbiased cross-modal understanding. Due to the lack of semantics, heterogeneous representations may lead to erroneous matches, especially in complex scenes with ambiguous visual content or interference from multiple audio sources. We introduce the multi-granularity implicit text (MIT) involving video-, segment- and frame-level as the bridge to establish the modality-shared space, reducing modality gaps and providing prior guidance. Visual content carries more information and typically dominates, thereby marginalizing audio features in the decision-making. To mitigate knowledge preference, we propose the semantic counterfactual (SC) to learn orthogonal representations in the latent space, generating diverse counterfactual samples, thus avoiding biases introduced by complex functional designs and explicit modifications of text structures or attributes. We further formulate the collaborative distribution-aware contrastive learning (CDCL), incorporating factual-counterfactual and inter-modality contrasts to align representations, promoting cohesion and decoupling. Extensive experiments on three public datasets validate that the proposed method achieves state-of-the-art performance.

Figures

Figures reproduced from arXiv: 2507.20740 by the authors.

Figure 1
Figure 1. The problem decomposition and method paradigm of AVS task. In the left subfigure, we divide the task into four quadrants based [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The framework of ICF. For any given audio-video pair, we obtain high-dimensional features using visual and audio encoders [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. t-SNE visualization. Drum (•), Marimba (•), Cello (•). pretrained on Youtube-8M as the audio backbone. To gen￾erate implicit orthogonal text, we employ Latent Diffusion [53] to reduce computational complexity. Following [76], we leverage Multi-Scale Deformable Attention Transformer as the pixel decoder. The loss balancing hyper-parameter settings and more experiments are included in the supple￾mentary material 1 . E… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison. From left to right, the samples are sourced from the S4, M3, and AVSS datasets. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Quantitative ablation of Top-K k c , orthogonality α o , number of diffusion steps s d , signal-to-noise ratio r a , and frame mixing r v . the performance gap to: 1) Inherent uncertainties (instabil￾ities) in audio, i.e., semantic biases caused by multiple fac￾tors (e…
Figure 6
Figure 6. Figure 6: Qualitative ablation of proposed components. From top [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

86 extracted references · 64 canonical work pages

  1. [1]

    Remembering the past and imagining the future: Common and distinct neural substrates during event construction and elaboration

    Donna Rose Addis, Alana T Wong, and Daniel L Schacter. Remembering the past and imagining the future: Common and distinct neural substrates during event construction and elaboration. Neuropsychologia, 45(7):1363–1377, 2007. 1

    2007

  2. [2]

    Unsupervised Audio-Visual Segmentation with Modality Alignment

    Swapnil Bhosale, Haosen Yang, Diptesh Kanojia, Jiangkang Deng, and Xiatian Zhu. Unsupervised audio-visual segmentation with modality alignment. arXiv preprint arXiv:2403.14203, 2024. 3

    work page Pith review arXiv 2024

  3. [3]

    Numerics of gram-schmidt orthogonalization

    ˚Ake Bj ¨orck. Numerics of gram-schmidt orthogonalization. Linear Algebra and Its Applications, 197:297–316, 1994. 4

    1994

  4. [4]

    Self-projection and the brain

    Randy L Buckner and Daniel C Carroll. Self-projection and the brain. Trends in cognitive sciences, 11(2):49–57, 2007. 1

    2007

  5. [5]

    Bootstrapping Audio-Visual Segmentation by Strengthening Audio Cues

    Tianxiang Chen, Zhentao Tan, Tao Gong, Qi Chu, Yue Wu, Bin Liu, Le Lu, Jieping Ye, and Nenghai Yu. Bootstrap- ping audio-visual segmentation by strengthening audio cues. arXiv preprint arXiv:2402.02327, 2024. 1, 4

    work page Pith review arXiv 2024

  6. [6]

    Unraveling in- stance associations: A closer look for audio-visual segmenta- tion

    Yuanhong Chen, Yuyuan Liu, Hu Wang, Fengbei Liu, Chong Wang, Helen Frazer, and Gustavo Carneiro. Unraveling in- stance associations: A closer look for audio-visual segmenta- tion. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 26497–26507,

  7. [7]

    Cpm: Class-conditional prompting ma- chine for audio-visual segmentation

    Yuanhong Chen, Chong Wang, Yuyuan Liu, Hu Wang, and Gustavo Carneiro. Cpm: Class-conditional prompting ma- chine for audio-visual segmentation. In European Confer- ence on Computer Vision, pages 438–456. Springer, 2024. 2, 3

    2024

  8. [8]

    Cpm: Class-conditional prompting ma- chine for audio-visual segmentation

    Yuanhong Chen, Chong Wang, Yuyuan Liu, Hu Wang, and Gustavo Carneiro. Cpm: Class-conditional prompting ma- chine for audio-visual segmentation. In European Confer- ence on Computer Vision , pages 438–456. Springer, 2025. 6

    2025

Show all 86 references
  1. [9]

    C-cam: Causal cam for weakly supervised seman- tic segmentation on medical image

    Zhang Chen, Zhiqiang Tian, Jihua Zhu, Ce Li, and Shaoyi Du. C-cam: Causal cam for weakly supervised seman- tic segmentation on medical image. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11676–11685, 2022. 3

    2022

  2. [10]

    Asi- seg: Audio-driven surgical instrument segmentation with surgeon intention understanding

    Zhen Chen, Zongming Zhang, Wenwu Guo, Xingjian Luo, Long Bai, Jinlin Wu, Hongliang Ren, and Hongbin Liu. Asi- seg: Audio-driven surgical instrument segmentation with surgeon intention understanding. In 2024 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems ...

    2024

  3. [11]

    Regret and its avoidance: a neuroimaging study of choice behavior

    Giorgio Coricelli, Hugo D Critchley, Mateus Joffily, John P O’Doherty, Angela Sirigu, and Raymond J Dolan. Regret and its avoidance: a neuroimaging study of choice behavior. Nature neuroscience, 8(9):1255–1262, 2005. 1

    2005

  4. [12]

    An image is worth one word: Personalizing text-to- image generation using textual inversion

    Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patash- nik, Amit H Bermano, Gal Chechik, and Daniel Cohen- Or. An image is worth one word: Personalizing text-to- image generation using textual inversion. arXiv preprint arXiv:2208.01618, 2022. 3

    2022 arXiv

  5. [13]

    Avsegformer: Audio-visual segmentation with trans- former

    Shengyi Gao, Zhe Chen, Guo Chen, Wenhai Wang, and Tong Lu. Avsegformer: Audio-visual segmentation with trans- former. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 12155–12163, 2024. 1, 3, 4, 6, 8

    2024

  6. [14]

    Open- vocabulary audio-visual semantic segmentation

    Ruohao Guo, Liao Qu, Dantong Niu, Yanyu Qi, Wenzhen Yue, Ji Shi, Bowei Xing, and Xianghua Ying. Open- vocabulary audio-visual semantic segmentation. In Proceed- ings of the 32nd ACM International Conference on Multime- dia, pages 7533–7541, 2024. 3

    2024

  7. [15]

    Embodied intelligence via learning and evolution

    Agrim Gupta, Silvio Savarese, Surya Ganguli, and Li Fei- Fei. Embodied intelligence via learning and evolution. Na- ture communications, 12(1):5721, 2021. 1

    2021

  8. [16]

    Improving audio-visual segmenta- tion with bidirectional generation

    Dawei Hao, Yuxin Mao, Bowen He, Xiaodong Han, Yuchao Dai, and Yiran Zhong. Improving audio-visual segmenta- tion with bidirectional generation. In Proceedings of the AAAI Conference on Artificial Intelligence , pages 2067– 2075, 2024. 1, 3, 6

    2024

  9. [17]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 5

    2016

  10. [18]

    Momentum contrast for unsupervised visual rep- resentation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 9729–9738, 2020. 5

    2020

  11. [19]

    Decoupling static and hier- archical motion perception for referring video segmentation

    Shuting He and Henghui Ding. Decoupling static and hier- archical motion perception for referring video segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13332–13341, 2024. 1

    2024

  12. [20]

    A general mechanism for perceptual decision-making in the human brain

    Hauke R Heekeren, Sean Marrett, Peter A Bandettini, and Leslie G Ungerleider. A general mechanism for perceptual decision-making in the human brain. Nature, 431(7010): 859–862, 2004. 3

    2004

  13. [21]

    Cnn archi- tectures for large-scale audio classification

    Shawn Hershey, Sourish Chaudhuri, Daniel PW Ellis, Jort F Gemmeke, Aren Jansen, R Channing Moore, Manoj Plakal, Devin Platt, Rif A Saurous, Bryan Seybold, et al. Cnn archi- tectures for large-scale audio classification. In 2017 ieee in- ternational conference on acoustics, spe...

    2017

  14. [22]

    Aleatory and epistemic uncertainty in prob- ability elicitation with an example from hazardous waste management

    Stephen C Hora. Aleatory and epistemic uncertainty in prob- ability elicitation with an example from hazardous waste management. Reliability Engineering & System Safety , 54 (2-3):217–223, 1996. 5

    1996

  15. [23]

    Mix and local- ize: Localizing sound sources in mixtures

    Xixi Hu, Ziyang Chen, and Andrew Owens. Mix and local- ize: Localizing sound sources in mixtures. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10483–10492, 2022. 2

    2022

  16. [24]

    Make-an-audio: Text-to-audio gen- eration with prompt-enhanced diffusion models

    Rongjie Huang, Jiawei Huang, Dongchao Yang, Yi Ren, Luping Liu, Mingze Li, Zhenhui Ye, Jinglin Liu, Xiang Yin, and Zhou Zhao. Make-an-audio: Text-to-audio gen- eration with prompt-enhanced diffusion models. In Inter- national Conference on Machine Learning , pages 13916– 13932...

    2023

  17. [25]

    Discovering sound- ing objects by audio queries for audio visual segmentation

    Shaofei Huang, Han Li, Yuqing Wang, Hongji Zhu, Jiao Dai, Jizhong Han, Wenge Rong, and Si Liu. Discovering sound- ing objects by audio queries for audio visual segmentation. arXiv preprint arXiv:2309.09501, 2023. 3, 6

    2023 arXiv

  18. [26]

    Adaptive mixtures of local experts.Neu- ral computation, 3(1):79–87, 1991

    Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of local experts.Neu- ral computation, 3(1):79–87, 1991. 8

    1991

  19. [27]

    Counterfactually augmented event matching for de-biased temporal sentence grounding

    Xun Jiang, Zhuoyuan Wei, Shenshen Li, Xing Xu, Jingkuan Song, and Heng Tao Shen. Counterfactually augmented event matching for de-biased temporal sentence grounding. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 6472–6481, 2024. 8

    2024

  20. [28]

    Learning to visually localize sound sources from mix- tures without prior source knowledge

    Dongjin Kim, Sung Jin Um, Sangmin Lee, and Jung Uk Kim. Learning to visually localize sound sources from mix- tures without prior source knowledge. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 26467–26476, 2024. 2

    2024

  21. [29]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. arXiv preprint arXiv:2304.02643, 2023. 6

    2023 arXiv

  22. [30]

    The singular value decompo- sition: Its computation and some applications

    Virginia Klema and Alan Laub. The singular value decompo- sition: Its computation and some applications. IEEE Trans- actions on automatic control, 25(2):164–176, 1980. 8

    1980

  23. [31]

    Improving vision and language concepts understanding with multimodal counterfactual samples

    Chengen Lai, Shengli Song, Sitong Yan, and Guangneng Hu. Improving vision and language concepts understanding with multimodal counterfactual samples. InEuropean Conference on Computer Vision, pages 174–191. Springer, 2024. 2

    2024

  24. [32]

    Selm: Selective mechanism based audio-visual segmentation

    Jiaxu Li, Songsong Yu, Yifan Wang, Lijun Wang, and Huchuan Lu. Selm: Selective mechanism based audio-visual segmentation. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 3926–3935, 2024. 3, 6

    2024

  25. [33]

    Catr: Combinatorial-dependence audio-queried transformer for audio-visual video segmentation

    Kexin Li, Zongxin Yang, Lei Chen, Yi Yang, and Jun Xiao. Catr: Combinatorial-dependence audio-queried transformer for audio-visual video segmentation. In Proceedings of the 31st ACM International Conference on Multimedia , pages 1485–1494, 2023. 3, 6

    2023

  26. [34]

    Dice loss for data-imbalanced nlp tasks

    Xiaoya Li, Xiaofei Sun, Yuxian Meng, Junjun Liang, Fei Wu, and Jiwei Li. Dice loss for data-imbalanced nlp tasks. arXiv preprint arXiv:1911.02855, 2019. 5

    1911 arXiv

  27. [35]

    Qdformer: Towards ro- bust audiovisual segmentation in complex environments with quantization-based semantic decomposition

    Xiang Li, Jinglu Wang, Xiaohao Xu, Xiulian Peng, Rita Singh, Yan Lu, and Bhiksha Raj. Qdformer: Towards ro- bust audiovisual segmentation in complex environments with quantization-based semantic decomposition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pa...

    2024

  28. [36]

    Bavs: bootstrapping audio- visual segmentation by integrating foundation knowledge

    Chen Liu, Peike Li, Hu Zhang, Lincheng Li, Zi Huang, Dadong Wang, and Xin Yu. Bavs: bootstrapping audio- visual segmentation by integrating foundation knowledge. IEEE Transactions on Multimedia, 2024. 2, 3, 6

    2024

  29. [37]

    Benchmarking au- dio visual segmentation for long-untrimmed videos

    Chen Liu, Peike Patrick Li, Qingtao Yu, Hongwei Sheng, Dadong Wang, Lincheng Li, and Xin Yu. Benchmarking au- dio visual segmentation for long-untrimmed videos. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22712–22722, 2024. 3

    2024

  30. [38]

    Pay attention to mlps

    Hanxiao Liu, Zihang Dai, David So, and Quoc V Le. Pay attention to mlps. Advances in neural information processing systems, 34:9204–9215, 2021. 4

    2021

  31. [39]

    Audio-aware query-enhanced trans- former for audio-visual segmentation

    Jinxiang Liu, Chen Ju, Chaofan Ma, Yanfeng Wang, Yu Wang, and Ya Zhang. Audio-aware query-enhanced trans- former for audio-visual segmentation. arXiv preprint arXiv:2307.13236, 2023. 1, 3, 6

    2023 arXiv

  32. [40]

    Annotation-free audio-visual segmentation

    Jinxiang Liu, Yu Wang, Chen Ju, Ya Zhang, and Weidi Xie. Annotation-free audio-visual segmentation. arXiv preprint arXiv:2305.11019, 2023. 3

    2023 arXiv

  33. [41]

    Audio-visual segmentation via unlabeled frame exploitation

    Jinxiang Liu, Yikun Liu, Fei Zhang, Chen Ju, Ya Zhang, and Yanfeng Wang. Audio-visual segmentation via unlabeled frame exploitation. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 26328–26339, 2024. 3, 6

    2024

  34. [42]

    Cross-modal causal relational reasoning for event-level visual question answer- ing

    Yang Liu, Guanbin Li, and Liang Lin. Cross-modal causal relational reasoning for event-level visual question answer- ing. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(10):11624–11641, 2023. 3

    2023

  35. [43]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021. 6

    2021

  36. [44]

    Step- ping stones: a progressive training strategy for audio-visual semantic segmentation

    Juncheng Ma, Peiwen Sun, Yaoting Wang, and Di Hu. Step- ping stones: a progressive training strategy for audio-visual semantic segmentation. In European Conference on Com- puter Vision, pages 311–327. Springer, 2024. 3

    2024

  37. [45]

    T-vsl: Text-guided visual sound source localization in mixtures

    Tanvir Mahmud, Yapeng Tian, and Diana Marculescu. T-vsl: Text-guided visual sound source localization in mixtures. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 26742–26751, 2024. 2

    2024

  38. [46]

    Contrastive conditional la- tent diffusion for audio-visual segmentation

    Yuxin Mao, Jing Zhang, Mochu Xiang, Yunqiu Lv, Yi- ran Zhong, and Yuchao Dai. Contrastive conditional la- tent diffusion for audio-visual segmentation. arXiv preprint arXiv:2307.16579, 2023. 3, 6

    2023 arXiv

  39. [47]

    Multimodal variational auto-encoder based audio-visual segmentation

    Yuxin Mao, Jing Zhang, Mochu Xiang, Yiran Zhong, and Yuchao Dai. Multimodal variational auto-encoder based audio-visual segmentation. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 954– 965, 2023. 3, 6

    2023

  40. [48]

    Weakly-supervised audio- visual segmentation

    Shentong Mo and Bhiksha Raj. Weakly-supervised audio- visual segmentation. Advances in Neural Information Pro- cessing Systems, 36:17208–17221, 2023. 3

    2023

  41. [49]

    Audio-visual grouping net- work for sound localization from mixtures

    Shentong Mo and Yapeng Tian. Audio-visual grouping net- work for sound localization from mixtures. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10565–10574, 2023. 2

    2023

  42. [50]

    The book of why: the new science of cause and effect

    Judea Pearl and Dana Mackenzie. The book of why: the new science of cause and effect. Basic books, 2018. 3

    2018

  43. [51]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  44. [52]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. In International confer- ence on machine learning, pages 8821–8831. Pmlr, 2021. 2

    2021

  45. [53]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 4, 6

    2022

  46. [54]

    Focal loss for dense ob- ject detection

    T-YLPG Ross and GKHP Doll ´ar. Focal loss for dense ob- ject detection. In proceedings of the IEEE conference on computer vision and pattern recognition, pages 2980–2988,

  47. [55]

    The wasserstein distance and approxi- mation theorems

    Ludger R ¨uschendorf. The wasserstein distance and approxi- mation theorems. Probability Theory and Related Fields, 70 (1):117–129, 1985. 5

    1985

  48. [56]

    Better aggregation in test-time augmentation

    Divya Shanmugam, Davis Blalock, Guha Balakrishnan, and John Guttag. Better aggregation in test-time augmentation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1214–1223, 2021. 5

    2021

  49. [57]

    A mathematical theory of commu- nication

    Claude Elwood Shannon. A mathematical theory of commu- nication. The Bell system technical journal, 27(3):379–423,

  50. [58]

    Looking similar sounding different: Leveraging counterfactual cross-modal pairs for audiovisual representa- tion learning

    Nikhil Singh, Chih-Wei Wu, Iroro Orife, and Mahdi Kalayeh. Looking similar sounding different: Leveraging counterfactual cross-modal pairs for audiovisual representa- tion learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 269...

    2024

  51. [59]

    3d audio-visual segmentation

    Artem Sokolov, Swapnil Bhosale, and Xiatian Zhu. 3d audio-visual segmentation. arXiv preprint arXiv:2411.02236, 2024. 3

    2024

  52. [60]

    Unveiling and mitigating bias in audio visual segmentation

    Peiwen Sun, Honggang Zhang, and Di Hu. Unveiling and mitigating bias in audio visual segmentation. In Proceedings of the 32nd ACM International Conference on Multimedia , pages 7259–7268, 2024. 3

    2024

  53. [61]

    Learning audio-visual source localization via false negative aware contrastive learning

    Weixuan Sun, Jiayi Zhang, Jianyuan Wang, Zheyuan Liu, Yiran Zhong, Tianpeng Feng, Yandong Guo, Yanhao Zhang, and Nick Barnes. Learning audio-visual source localization via false negative aware contrastive learning. In Proceed- ings of the IEEE/CVF Conference on Computer Vision...

    2023

  54. [62]

    Language-guided audio-visual source separation via trimodal consistency

    Reuben Tan, Arijit Ray, Andrea Burns, Bryan A Plum- mer, Justin Salamon, Oriol Nieto, Bryan Russell, and Kate Saenko. Language-guided audio-visual source separation via trimodal consistency. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition...

    2023

  55. [63]

    Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023. 6

    2023 arXiv

  56. [64]

    Neural discrete representation learning

    Aaron Van Den Oord, Oriol Vinyals, et al. Neural discrete representation learning. Advances in neural information pro- cessing systems, 30, 2017. 8

    2017

  57. [65]

    Counterfactual cycle-consistent learn- ing for instruction following and generation in vision- language navigation

    Hanqing Wang, Wei Liang, Jianbing Shen, Luc Van Gool, and Wenguan Wang. Counterfactual cycle-consistent learn- ing for instruction following and generation in vision- language navigation. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , ...

    2022

  58. [66]

    Vision-and-language naviga- tion via causal learning

    Liuyi Wang, Zongtao He, Ronghao Dang, Mengjiao Shen, Chengju Liu, and Qijun Chen. Vision-and-language naviga- tion via causal learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 13139–13150, 2024. 3

    2024

  59. [67]

    Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, et al. Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution. arXiv preprint arXiv:2409.12191, 2024. 8

    2024 arXiv

  60. [68]

    Pvt v2: Improved baselines with pyramid vision transformer

    Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pvt v2: Improved baselines with pyramid vision transformer. Computational Visual Media, 8(3):415–424, 2022. 5

    2022

  61. [69]

    Drivedreamer: Towards real-world- drive world models for autonomous driving

    Xiaofeng Wang, Zheng Zhu, Guan Huang, Xinze Chen, Jia- gang Zhu, and Jiwen Lu. Drivedreamer: Towards real-world- drive world models for autonomous driving. In European Conference on Computer Vision , pages 55–72. Springer,

  62. [70]

    Prompting segmentation with sound is generalizable audio-visual source localizer

    Yaoting Wang, Weisong Liu, Guangyao Li, Jian Ding, Di Hu, and Xi Li. Prompting segmentation with sound is generalizable audio-visual source localizer. arXiv preprint arXiv:2309.07929, 2023. 3

    2023 arXiv

  63. [71]

    Prompting segmentation with sound is gen- eralizable audio-visual source localizer

    Yaoting Wang, Weisong Liu, Guangyao Li, Jian Ding, Di Hu, and Xi Li. Prompting segmentation with sound is gen- eralizable audio-visual source localizer. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 5669– 5677, 2024. 3

    2024

  64. [72]

    Can textual semantics mitigate sounding object segmentation preference? In European Conference on Com- puter Vision, pages 340–356

    Yaoting Wang, Peiwen Sun, Yuanchao Li, Honggang Zhang, and Di Hu. Can textual semantics mitigate sounding object segmentation preference? In European Conference on Com- puter Vision, pages 340–356. Springer, 2024. 2, 6

    2024

  65. [73]

    Large-scale con- trastive language-audio pretraining with feature fusion and keyword-to-caption augmentation

    Yusong Wu, Ke Chen, Tianyu Zhang, Yuchen Hui, Taylor Berg-Kirkpatrick, and Shlomo Dubnov. Large-scale con- trastive language-audio pretraining with feature fusion and keyword-to-caption augmentation. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Si...

    2023

  66. [74]

    Videoclip: Contrastive pre-training for zero-shot video-text understanding

    Hu Xu, Gargi Ghosh, Po-Yao Huang, Dmytro Okhonko, Armen Aghajanyan, Florian Metze, Luke Zettlemoyer, and Christoph Feichtenhofer. Videoclip: Contrastive pre-training for zero-shot video-text understanding. arXiv preprint arXiv:2109.14084, 2021. 3

    2021 arXiv

  67. [75]

    Referred by multi-modality: A unified tem- poral transformer for video object segmentation

    Shilin Yan, Renrui Zhang, Ziyu Guo, Wenchao Chen, Wei Zhang, Hongyang Li, Yu Qiao, Hao Dong, Zhongjiang He, and Peng Gao. Referred by multi-modality: A unified tem- poral transformer for video object segmentation. InProceed- ings of the AAAI Conference on Artificial Intelligen...

    2024

  68. [76]

    Cooperation does matter: Exploring multi-order bilateral relations for audio- visual segmentation

    Qi Yang, Xing Nie, Tong Li, Pengfei Gao, Ying Guo, Cheng Zhen, Pengfei Yan, and Shiming Xiang. Cooperation does matter: Exploring multi-order bilateral relations for audio- visual segmentation. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognit...

    2024

  69. [77]

    Revisiting counterfactual prob- lems in referring expression comprehension

    Zhihan Yu and Ruifan Li. Revisiting counterfactual prob- lems in referring expression comprehension. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 13438–13448, 2024. 2

    2024

  70. [78]

    Losh: Long-short text joint prediction network for referring video object segmentation

    Linfeng Yuan, Miaojing Shi, Zijie Yue, and Qijun Chen. Losh: Long-short text joint prediction network for referring video object segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 14001–14010, 2024. 1

    2024

  71. [79]

    Discovering the real association: Multimodal causal rea- soning in video question answering

    Chuanqi Zang, Hanqing Wang, Mingtao Pei, and Wei Liang. Discovering the real association: Multimodal causal rea- soning in video question answering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19027–19036, 2023. 3

    2023

  72. [80]

    Weakly- supervised mirror detection via scribble annotations

    Mingfeng Zha, Yunqiang Pei, Guoqing Wang, Tianyu Li, Yang Yang, Wenbin Qian, and Heng Tao Shen. Weakly- supervised mirror detection via scribble annotations. In Pro- ceedings of the AAAI conference on artificial intelligence , pages 6953–6961, 2024. 3

    2024

  73. [81]

    Heterogeneous experts and hierarchical perception for un- derwater salient object detection

    Mingfeng Zha, Guoqing Wang, Yunqiang Pei, Tianyu Li, Xiongxin Tang, Chongyi Li, Yang Yang, and Heng Tao Shen. Heterogeneous experts and hierarchical perception for un- derwater salient object detection. IEEE Transactions on Im- age Processing, 2025. 3

    2025

  74. [82]

    Causal intervention for weakly- supervised semantic segmentation

    Dong Zhang, Hanwang Zhang, Jinhui Tang, Xian-Sheng Hua, and Qianru Sun. Causal intervention for weakly- supervised semantic segmentation. Advances in Neural In- formation Processing Systems, 33:655–666, 2020. 3

    2020

  75. [83]

    Audio–visual segmentation

    Jinxing Zhou, Jianyuan Wang, Jiayi Zhang, Weixuan Sun, Jing Zhang, Stan Birchfield, Dan Guo, Lingpeng Kong, Meng Wang, and Yiran Zhong. Audio–visual segmentation. In European Conference on Computer Vision , pages 386–

  76. [84]

    Audio-visual segmentation with semantics

    Jinxing Zhou, Xuyang Shen, Jianyuan Wang, Jiayi Zhang, Weixuan Sun, Jing Zhang, Stan Birchfield, Dan Guo, Ling- peng Kong, Meng Wang, et al. Audio-visual segmentation with semantics. International Journal of Computer Vision , pages 1–21, 2024. 5, 6

    2024

  77. [85]

    Exploring pre-trained text- to-video diffusion models for referring video object segmen- tation

    Zixin Zhu, Xuelu Feng, Dongdong Chen, Junsong Yuan, Chunming Qiao, and Gang Hua. Exploring pre-trained text- to-video diffusion models for referring video object segmen- tation. In European Conference on Computer Vision, pages 452–469. Springer, 2024. 1

    2024

  78. [403]

    2, 5, 6, 8

    Springer, 2022. 2, 5, 6, 8

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.