Pith. sign in

REVIEW 2 major objections 4 minor 57 references

AeroReformer2: Spoken-Query Referring Segmentation for Aerial Images

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read A new benchmark and model establish that spoken, full-sentence queries can drive referring segmentation in aerial imagery, with the Swin-Base model reaching 62.09% clean mIoU.

desk verdict Solid new benchmark and model for spoken-query aerial referring segmentation, but the closed-set TTS voice protocol means the reported gains don't measure speaker generalization; the paper deserves review with a request for a leave-one-voice-out split. read the letter →

arxiv 2608.08874 v2 pith:XJU43JDP submitted 2026-08-09 cs.CV

classification cs.CV
keywords spoken-queryreferringsegmentationremotesensingimageryimagespeech-conditionedcross-modalattentionlinearaudiorobustnessaerialanalysis
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

This paper tries to establish that spoken, full-sentence referring expressions can be a usable, hands-free interface for pixel-level remote-sensing analysis, and that the difference between written and spoken grounding deserves its own benchmark. To that end it builds VoiceAeroRef, a 52,466-triplet spoken-query benchmark derived from RISBench, with eight accent–gender text-to-speech voices for training and a nine-condition hard test set that mixes synthesized rotor, wind, and mixed interference at 15, 7.5, and 0 dB. It also proposes AeroReformer2, a bilateral network that preserves the speech token sequence through a confidence-gated memory and dual-scale kernel linear cross-attention instead of materializing a dense speech–pixel affinity matrix. On the clean test split, the Swin-Base version reports 62.09% mean intersection over union (mIoU) and 68.22% overall intersection over union (oIoU), exceeding the strongest audio-adapted baseline by 5.38 and 2.08 percentage points, respectively, and it keeps the best hard-set mIoU at 54.09%. If these numbers hold, the benchmark gives the community a reusable way to separate acoustic robustness from visual grounding quality.

What carries the argument

The load-bearing machinery is the pair (VoiceAeroRef, AeroReformer2). VoiceAeroRef supplies the controlled speech protocol: eight accent–gender neural TTS renderings per training expression with one voice sampled per epoch, balanced single-voice clean evaluation, and a nine-condition hard grid of synthesized rotor, wind, and mixed noise at 15, 7.5, and 0 dB. AeroReformer2's core operator is speech–visual kernel linear attention (SV-KLA), which rewrites cross-attention associatively as $\phi(Q)(\phi(K)^\top U)$ so that each visual query conditions on a fixed $d\times d$ speech memory instead of an $N\times T$ affinity matrix; a confidence-gated token memory (CGTM) weights each speech token before that accumulation. Around this operator, a bilateral visual backbone keeps a detail path at 1/8 resolution for small-object boundaries, fuses speech at 1/16 and 1/8 scales for coarse-to-fine grounding, and a 1/4 refinement head restores fine boundaries after fusion. The whole design is meant to let a variable-length utterance ground repeated, tiny, and arbitrarily oriented aerial targets without quadratic memory in the token count.

What would settle it

Record the same 16,159 test expressions spoken by human analysts, with and without field-recorded rotor and wind noise, run the released AeroReformer2 checkpoints on them, and compare clean and hard mIoU with the reported 62.09% and 54.09%; a material drop would show that the synthetic speech and noise protocol overstates real-world robustness.

Watch

Extended reading notes

Core claim

The paper's central claim is that full-sentence spoken-query referring segmentation for remote-sensing imagery is a distinct, benchmarkable task, and that it is solvable well beyond pooled-audio conditioning. The authors claim that VoiceAeroRef is the first benchmark of this kind: it preserves the original RISBench image–expression–mask triplets, adds controlled speaker diversity, and adds a hard evaluation split in which rotor, wind, and mixed interference are applied at three signal-to-noise levels. They further claim that AeroReformer2-Swin-B is the current state of the art on that benchmark, with 62.09% clean mIoU versus 56.71% for the strongest adapted text-based RRSIS model, and 54.09% hard mIoU versus 48.46%. The paper interprets this gap as evidence that token-preserving speech fusion, rather than sentence-level conditioning, is what lets relational phrases such as “partially visible at the lower edge” or “slightly left from the center” guide segmentation.

Load-bearing premise

The load-bearing premise is that synthesized speech from eight neural text-to-speech voices and locally generated rotor and wind noise behave enough like real analyst commands and field interference; if real speech or real noise differs acoustically, the measured clean and hard accuracy will not transfer to deployment.

Editorial extensions

If this is right

  • Text-only RRSIS models can now be compared with spoken-query variants on the same images, masks, and splits, so the cost of acoustic encoding becomes isolable from visual grounding quality.
  • The nine-condition hard grid turns acoustic robustness into a measurable axis: a model can be scored per noise type and SNR, and checkpoints chosen on clean validation still receive a hard-set number.
  • Because SV-KLA never materializes an $N\times T$ affinity matrix, the fusion design can scale to longer utterances or higher-resolution feature maps without quadratic memory growth.
  • The reported margins over audio-adapted text models suggest that keeping the token sequence through fusion, rather than pooling to a sentence vector, is the main architectural lever.
  • The split policy keeps the training set at 26,294 effective samples per epoch despite eight voice recordings per source, so speaker diversity does not inflate epoch size.

Reading between the lines

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

  • A natural next test is to replace the eight TTS voices with recordings from real analysts; the paper's own future-work list flags this, and a large drop would indicate that synthesized speech overstates clean-set transferability.
  • The hard-noise synthesis is a controlled proxy, not a field model: real rotor and wind spectra, microphone placement, and radio-channel effects could shift the 0 dB results, so deployment claims should be read as upper bounds until matched conditions are tested.
  • The learned confidence gate could be repurposed beyond segmentation: if token weights correlate with acoustic corruption, they could drive uncertainty estimates for the whole mask or trigger a clarification dialog when the query is ambiguous.
  • The same dual-scale linear-attention routing could transfer to streaming speech, where the fixed-size $d\times d$ memory makes incremental token updates natural.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper introduces VoiceAeroRef, a spoken-query referring segmentation benchmark built from RISBench by converting every expression into speech with eight neural TTS voices (four accents x two genders), preserving the original image/mask splits, and adding a hard test set in which clean test utterances are mixed with locally synthesized rotor, wind, or mixed noise at three SNR levels. The paper also proposes AeroReformer2, a bilateral visual encoder with wav2vec 2.0 speech tokens, a confidence-gated token memory (CGTM), dual-scale kernel linear cross-modal attention (SV-KLA), and a high-resolution refinement head. On the clean test split, AeroReformer2-Swin-B achieves 62.09% mIoU and 68.22% oIoU, and on the hard split 54.09% mIoU, with margins of roughly 5.4-5.6 points over the strongest adapted baselines. The paper includes detailed experimental comparisons, ablations, efficiency measurements, and analytical complexity calculations.

Significance. If the closed-set voice protocol is addressed, the benchmark fills a genuine gap: no existing RRSIS resource provides spoken queries, and the hard-noise grid is a useful stress test. The model design is technically sound: the kernel-linear formulation avoids materializing an N x T affinity matrix, and the reported efficiency numbers in Table 8 and Table 9 are consistent with the stated equations. The paper is also honest about missing ablations (1/4 head) and the synthetic nature of the speech and noise. However, the lack of any held-out voice condition means the reported state-of-the-art numbers are within-voice-set results; this is the main threat to the central claim and must be fixed before the benchmark can be considered a speech-robustness testbed.

major comments (2)
  1. [Section 3.3, Table 3] All training, validation, and test utterances are synthesized from the same eight TTS voices; validation and test only re-assign one of those voices to each expression, and the hard set reuses the same clean test voice assignments. The reported 62.09% clean mIoU and 54.09% hard mIoU therefore measure adaptation to eight known voices, not generalization to unseen speakers, and the 5.38-point margin over the best baseline could shrink or reverse under a leave-one-voice-out evaluation. Please add an unseen-voice evaluation (e.g., train on seven voices and test on the eighth, reported for all models), and if VoiceAeroRef is to serve as a speech-robustness benchmark, make such a split an official protocol element.
  2. [Section 5.4, Table 7] Section 5.2 attributes the high-overlap gains (Pr@0.8 and Pr@0.9) to the 1/4 refinement head, but Section 5.4 states that 'A matched ablation of the 1/4 head is not available in the current experiment set.' Without an ablation that removes only this head, the contribution of the refinement head to the reported gains is unquantified and the attribution is unsupported. Please add the missing ablation and, if the head is retained, report the corresponding metric changes (including the effect on Pr@0.8/0.9).
minor comments (4)
  1. [Table 8] The AeroReformer2-Swin-B row is formatted incorrectly in the manuscript (the parameter and latency values are run together, e.g., '87.69239.19'); please fix the table layout so all columns are legible.
  2. [Section 5.4] No multi-seed or error-bar results are reported for any method. Because the key margins over baselines are a few mIoU points, a small repeated-run analysis (e.g., three seeds for the main models) would substantially strengthen the claims.
  3. [Abstract / Section 5.1] The abstract promises public release of code, but no URL or repository information is given and the paper does not state how the benchmark data (including the synthesized audio) will be distributed; please provide a concrete availability statement.
  4. [Section 3.4 / Section 6] The hard test relies solely on locally synthesized rotor, wind, and mixed noise, and the paper defers real analyst speech and field interference to future work. This is an honest limitation, but the paper should state explicitly in the conclusions (and ideally in the abstract) that the reported hard-set robustness numbers apply to synthetic interference superimposed on the same eight TTS voices, not to field conditions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark and architecture are evaluated on a held-out test split, and prior self-citations serve only as background building blocks.

full rationale

I walked the claimed derivation chain: (i) VoiceAeroRef is constructed from RISBench by TTS synthesis while preserving image/mask/splits; (ii) the hard test set reuses the clean test voice assignments under synthetic interference; (iii) AeroReformer2's equations (3)-(14) form a standard kernelized linear-attention formulation with a learned confidence gate and bilateral refinement; and (iv) the baselines are audio-adaptations of published text-guided RRSIS models sharing the same wav2vec 2.0 tokens. The held-out test/validation protocol is independent of training: checkpoints are selected on validation mIoU (Section 5.1), and no hard-set fine-tuning is performed (Section 5.3). The authors' prior work (ABCNet, LSwinSR, AeroReformer) is cited as background or building-block context, not as a load-bearing uniqueness theorem; the bilateral and linear-attention ingredients are also independently established by external citations (BiSeNet, Efficient Attention, Katharopoulos et al.). The one legitimate limitation—the eight TTS voices used at test are drawn from the same voice categories as training, so the results do not measure unseen-speaker generalization—is an external-validity and protocol gap, not a circular reduction: no reported mIoU is equal to a training input by construction. Thus no circular step is exhibited, and the paper is self-contained against an externally specified test split.

Assumptions & free parameters 4 free parameters · 5 assumptions · 2 invented entities

The central claims rest on the source benchmark, the TTS proxy, the noise synthesis, and the frozen speech encoder; these are domain assumptions rather than standard math. The listed free parameters are hand-chosen evaluation design values, not fitted physical constants.

free parameters (4)
  • Hard-test SNR levels = 15, 7.5 and 0 dB plus or minus 1.5 dB jitter
    Hand-selected severity grid defining the low, medium, and high hard-test conditions in Section 3.4; the hard-set results in Tables 5 and 6 depend on these values.
  • Mixed-noise composition ratio = 0.6 rotor / 0.4 wind
    Hand-chosen mixture for the mixed interference condition in Section 3.4; no justification is given for this specific ratio.
  • Audio duration cap = 8 seconds
    Utterances are capped at 8 seconds before wav2vec 2.0 encoding, which affects the token count and the efficiency profile in the experimental protocol.
  • Training resolution = 352 by 352 pixels
    All models are trained at 352 by 352 and evaluated at 512 by 512; this is a hand-chosen design point that affects accuracy and memory.
assumptions (5)
  • domain assumption RISBench image-expression-mask triplets are correct and suitable as source data.
    VoiceAeroRef preserves the images, masks, and splits of RISBench; any label noise or ambiguous expressions in RISBench are inherited. See Section 3.2.
  • domain assumption Neural TTS voices are a sufficient proxy for real accented and gendered speech.
    The claim that the benchmark measures spoken-query robustness depends on synthetic voices capturing the acoustic variability of real commands. See Section 3.3 and the future-work acknowledgment in Section 6.
  • domain assumption Synthesized rotor and wind noise models reproduce field interference faithfully.
    Hard-test results are claimed under rotor, wind, and mixed interference, but the noise is locally synthesized and does not model the acquisition platform. See Section 3.4.
  • domain assumption Frozen wav2vec 2.0 features provide task-relevant speech tokens.
    The entire fusion path operates on wav2vec 2.0 outputs without transcripts or ASR; if these features obscure referring words, the benchmark results would not reflect the true task. See Section 4.
  • standard math The associative property of matrix products permits the linear-attention rearrangement.
    Equations (6) and (7) rely on the standard kernel-linear-attention rearrangement from Katharopoulos et al. [33]. This is a standard mathematical identity.
invented entities (2)
  • VoiceAeroRef
    purpose: Spoken-query referring segmentation benchmark for remote-sensing imagery, derived from RISBench through TTS synthesis.
    The paper provides no release URL or downloadable archive; the benchmark exists only as a construction description, so others cannot yet verify or extend it.
  • AeroReformer2
    purpose: Speech-conditioned bilateral segmentation network with confidence-gated token memory and dual-scale speech-visual kernel linear attention.
    No code or checkpoints are provided; the architecture's behavior is evidenced only by the paper's own tables, with no independently reproducible artifacts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AeroReformer2: Spoken-Query Referring Segmentation for Aerial Images." pith.science (2026). https://pith.science/paper/XJU43JDP

@misc{pith2026260808874,
  author       = {Pith},
  title        = {Pith review of: AeroReformer2: Spoken-Query Referring Segmentation for Aerial Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XJU43JDP}},
  note         = {Machine review of arXiv:2608.08874}
}
read the original abstract

Spoken language offers a natural, hands-free interface for specifying an arbitrary target in dense remote-sensing imagery, yet existing referring remote-sensing image segmentation benchmarks accept only written expressions. To bridge this gap, we introduce \dataset, a spoken-query benchmark derived from RISBench that adds accent- and voice-diverse speech while preserving the original image, mask, and data splits. Its hard evaluation sets combine rotor, wind, and mixed interference with three signal-to-noise levels. We also propose \model, an efficient bilateral network that combines a boundary-preserving visual path with token-preserving speech encoding, kernel linear cross-modal attention, and a resolution refinement head. The design conditions visual features at two scales without materializing a dense speech--visual affinity matrix, then restores fine boundaries using high-resolution visual features. On the clean test split, \model with Swin-Base achieves 62.09\% mean intersection over union (mIoU) and 68.22\% overall intersection over union (oIoU), outperforming the strongest audio-adapted remote-sensing baseline by 5.38 and 2.08 percentage points, respectively. It retains the best hard-set mIoU at 54.09\%. To the best of our knowledge, this is the first benchmark and model study of full-sentence spoken-query referring segmentation for remote-sensing imagery. The code will be made publicly available.

Figures

Figures reproduced from arXiv: 2608.08874 by the authors.

Figure 1
Figure 1. VoiceAeroRef dataset construction and evaluation protocol. The pipeline covers source-data quality control, controlled neural speech synthesis, hard-test construction, and split-specific voice sampling. 3 6 9 12 15 18 21 24 27 30 33 36 0 5 10 15 Words per expression Expressions (%) (a) Expression length 0 20 40 60 80 Orientation Quantity Shape Color/tone Relative relation Size/extent Position/direction Expressions c… view at source ↗
Figure 2
Figure 2. Linguistic characteristics of the usable expressions. Panel (a) gives the frequency of each expression length from 3 to 36 words; only five expressions are longer than 36 words. Panel (b) reports the prevalence of lexicon-based linguistic cues. Cue categories are non-exclusive and therefore do not sum to 100%. 2. Related Work 2.1. Referring image and remote-sensing segmentation RIS was introduced as pixel-level grou… view at source ↗
Figure 3
Figure 3. Lexical and referent-semantic composition of the 52,466 usable expressions. Panel (a) shows the 100 most frequent non-stopword tokens, with font size proportional to log frequency. Panel (b) is a Nightingale rose of first-mentioned referent families; sector area is proportional to expression share. “Other” combines buildings, natural/water terms, and unmatched expressions [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: AeroReformer2 architecture for spoken-query aerial referring segmentation. The model combines bilateral visual encoding, confidence-gated speech representations, dual-scale speech–visual kernel linear attention, and high-resolution mask refinement. The speech path tran…
Figure 5
Figure 5. Figure 5: Baseball-field example. The instruction selects the smaller field that is only partially visible at the lower image boundary. AeroReformer2 recovers that truncated diamond with 0.971 IoU, whereas all five comparison methods select the fully visible central field and th…
Figure 6
Figure 6. Figure 6: Small-object example. The instruction refers to the object slightly left of image center. AeroReformer2 isolates the compact target with 0.878 IoU, whereas the alternatives activate on larger structures, image boundaries, or multiple distractors. European Conference on…
Figure 7
Figure 7. Figure 7: Waterfront-vehicle example. The target is a very small vehicle on the far-left side of the waterfront. AeroReformer2 preserves the correct location and compact extent with 0.875 IoU; the comparison methods miss it, fragment it, or select a different object. mentation, …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 44 canonical work pages

  1. [1]

    Zhang, S

    Q. Zhang, S. Zheng, C. Zhang, X. Wang, R. Li, Efficient large-scale oblique image matching based on cascade hash- ing and match data scheduling, Pattern Recognition 138 (2023) 109442

  2. [2]

    R. Li, X. Zhao, AeroReformer: Aerial referring trans- former for UA V-based referring image segmentation, Inter- national Journal of Applied Earth Observation and Geoin- formation 143 (2025) 104817. doi:10.1016/j.jag. 2025.104817

  3. [3]

    R. Li, J. Zhang, X. Zhao, Long-range near-surface wake signatures of offshore wind farm clusters revealed by satellite observations, Communications Engineeringdoi: 10.1038/s44172-026-00684-7

  4. [4]

    J. Long, E. Shelhamer, T. Darrell, Fully convolutional networks for semantic segmentation, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, pp. 3431–3440

  5. [5]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-Net: Convolutional networks for biomedical image segmentation, in: Medical Image Computing and Computer-Assisted Intervention, 2015, pp. 234–241

  6. [6]

    L.-C. Chen, G. Papandreou, F. Schroff, H. Adam, Rethink- ing atrous convolution for semantic image segmentation, arXiv preprint arXiv:1706.05587

  7. [7]

    R. Li, X. Zhao, LSwinSR: UA V imagery super-resolution based on linear Swin Transformer, IEEE Transactions on Geoscience and Remote Sensing 62 (2024) 1–13. doi: 10.1109/TGRS.2024.3463204

  8. [8]

    Z. Yuan, L. Mou, Y . Hua, X. X. Zhu, RRSIS: Referring remote sensing image segmentation, IEEE Transactions on Geoscience and Remote Sensing 62 (2024) 1–16. doi: 10.1109/TGRS.2024.3357333

Show all 57 references
  1. [9]

    S. Liu, Y . Ma, X. Zhang, H. Wang, J. Ji, X. Sun, R. Ji, Rotated multi-scale interaction network for referring re- mote sensing image segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 26658–26668

  2. [10]

    Z. Dong, Y . Sun, Y . Gu, T. Liu, Cross-modal bidirectional interaction model for referring remote sensing image seg- mentation, arXiv preprint arXiv:2410.08613. URLhttps://arxiv.org/abs/2410.08613

  3. [11]

    Z. Yang, H. Yao, L. Tian, X. Zhao, Q. Li, Q. Wang, A large-scale referring remote sensing image segmentation dataset and benchmark, arXiv preprint arXiv:2506.03583. URLhttps://arxiv.org/abs/2506.03583

  4. [12]

    J. Zhou, J. Wang, J. Zhang, W. Sun, J. Zhang, S. Birchfield, D. Guo, L. Kong, M. Wang, Y . Zhong, Audio-visual seg- mentation, in: European Conference on Computer Vision, 2022

  5. [13]

    The object located slightly left from the center of the image

    Y . Wang, P. Sun, D. Zhou, G. Li, H. Zhang, D. Hu, Ref- A VS: Refer and segment objects in audio-visual scenes, in: 12 Input RGB Ground truth AeroReformer2 (ours) IoU 0.878 LSCF adapter IoU 0.084 FIANet adapter IoU 0.000 SBANet adapter IoU 0.102 DeepLabV3-R101 IoU 0.000 FCN-R1...

  6. [14]

    W. Pan, H. Shi, Z. Zhao, J. Zhu, X. He, Z. Pan, L. Gao, J. Yu, F. Wu, Q. Tian, Wnet: Audio-guided video object segmentation via wavelet-based cross-modal denoising net- works, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 1320– 1331

  7. [15]

    J. A. Santos, Z. Wu, X. Alameda-Pineda, R. Timofte, Lay- over or direct flight: Rethinking audio-guided image seg- mentation, arXiv preprint arXiv:2511.22025

  8. [16]

    K. Ying, H. Ding, G. Jie, Y .-G. Jiang, Towards omnimodal expressions and reasoning in referring audio-visual seg- mentation, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 22575–22585

  9. [17]

    R. Hu, M. Rohrbach, T. Darrell, Segmentation from nat- ural language expressions, in: European Conference on Computer Vision, 2016, pp. 108–124

  10. [18]

    C. Liu, Z. Lin, X. Shen, J. Yang, X. Lu, A. Yuille, Recur- rent multimodal interaction for referring image segmenta- tion, in: Proceedings of the IEEE International Conference on Computer Vision, 2017, pp. 1271–1280

  11. [19]

    R. Li, K. Li, Y .-C. Kuo, M. Shu, X. Qi, X. Shen, J. Jia, Referring image segmentation via recurrent refinement networks, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 5745– 5753

  12. [20]

    L. Ye, M. Rochan, Z. Liu, Y . Wang, Cross-modal self- attention network for referring image segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 10502–10511

  13. [21]

    Huang, T

    S. Huang, T. Hui, S. Liu, G. Li, Y . Wei, J. Han, L. Liu, B. Li, Referring image segmentation via cross-modal pro- gressive comprehension, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 10488–10497

  14. [22]

    Z. Yang, J. Wang, Y . Tang, K. Chen, H. Zhao, P. H. S. Torr, LA VT: Language-aware vision transformer for referring image segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 18155–18165

  15. [23]

    Z. Wang, Y . Lu, Q. Li, X. Tao, Y . Guo, M. Gong, T. Liu, CRIS: CLIP-driven referring image segmentation, in: Pro- ceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, 2022, pp. 11686–11695

  16. [24]

    The small vehicle located on the far left side of the waterfront area

    Y . Hu, Q. Wang, W. Shao, E. Xie, Z. Li, J. Han, P. Luo, Beyond one-to-one: Rethinking the referring image seg- 13 Input RGB Ground truth AeroReformer2 (ours) IoU 0.875 LSCF adapter IoU 0.000 FIANet adapter IoU 0.249 SBANet adapter IoU 0.084 DeepLabV3-R101 IoU 0.000 FCN-R101 I...

  17. [25]

    N. A. Shah, V . VS, V . M. Patel, LQMFormer: Language- aware query mask transformer for referring image seg- mentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 12903–12913

  18. [26]

    G.-S. Xia, X. Bai, J. Ding, Z. Zhu, S. Belongie, J. Luo, M. Datcu, M. Pelillo, L. Zhang, DOTA: A large-scale dataset for object detection in aerial images, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2018) 3974–3983

  19. [27]

    K. Li, G. Wan, G. Cheng, L. Meng, J. Han, Object detection in optical remote sensing images: A survey and a new benchmark, in: ISPRS Journal of Photogrammetry and Remote Sensing, V ol. 159, 2020, pp. 296–307

  20. [28]

    X. Li, J. Wang, X. Xu, M. Yang, F. Yang, Y . Zhao, R. Singh, B. Raj, Towards noise-tolerant speech-referring video object segmentation: Bridging speech and text, in: Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, Association for Computa...

  21. [29]

    C. Yu, J. Wang, C. Peng, C. Gao, G. Yu, N. Sang, BiSeNet: Bilateral segmentation network for real-time semantic seg- mentation, in: European Conference on Computer Vision, 2018, pp. 325–341

  22. [30]

    R. Li, S. Zheng, C. Zhang, C. Duan, L. Wang, P. M. Atkinson, ABCNet: Attentive bilateral contextual network for efficient semantic segmentation of fine-resolution re- motely sensed imagery, in: ISPRS Journal of Photogram- metry and Remote Sensing, V ol. 181, 2021, pp. 84–98. d...

  23. [31]

    M. Fan, S. Lai, J. Huang, X. Wei, Z. Chai, J. Luo, X. Wei, Rethinking BiSeNet for real-time semantic segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 9716–9725

  24. [32]

    Z. Shen, M. Zhang, H. Zhao, S. Yi, H. Li, Efficient atten- tion: Attention with linear complexities, in: IEEE Winter Conference on Applications of Computer Vision, 2021, pp. 3531–3539

  25. [33]

    Katharopoulos, A

    A. Katharopoulos, A. Vyas, N. Pappas, F. Fleuret, Trans- formers are RNNs: Fast autoregressive transformers with linear attention, in: Proceedings of the International Con- ference on Machine Learning, 2020, pp. 5156–5165

  26. [34]

    Baevski, Y

    A. Baevski, Y . Zhou, A. Mohamed, M. Auli, wav2vec 2.0: A framework for self-supervised learning of speech repre- 14 sentations, in: Advances in Neural Information Processing Systems, V ol. 33, 2020, pp. 12449–12460

  27. [35]

    T. Hui, S. Liu, S. Huang, G. Li, S. Yu, F. Zhang, J. Han, Linguistic structure guided context modeling for referring image segmentation, in: European Conference on Com- puter Vision, 2020, pp. 59–75

  28. [36]

    S. Yu, P. H. Seo, J. Son, Zero-shot referring image segmen- tation with global–local context features, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 19456–19465

  29. [37]

    Kirillov, E

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, et al., Segment anything, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 4015–4026

  30. [38]

    Shang, Z

    C. Shang, Z. Song, H. Qiu, L. Wang, F. Meng, H. Li, Prompt-driven referring image segmentation with instance contrasting, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 4124–4134

  31. [39]

    Y . X. Chng, H. Zheng, Y . Han, X. Qiu, G. Huang, Mask grounding for referring image segmentation, in: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 26573–26583

  32. [40]

    Ito, Feature design for bridging SAM and CLIP to- ward referring image segmentation, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Com- puter Vision, 2025

    K. Ito, Feature design for bridging SAM and CLIP to- ward referring image segmentation, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Com- puter Vision, 2025

  33. [41]

    T. Liu, S. Li, Hybrid global–local representation with aug- mented spatial guidance for zero-shot referring image seg- mentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 29634–29643

  34. [42]

    X. Li, J. Ding, M. Elhoseiny, VRSBench: A versatile vision-language benchmark dataset for remote sensing image understanding, in: Advances in Neural Informa- tion Processing Systems, Datasets and Benchmarks Track, 2024. URLhttps://arxiv.org/abs/2406.12384

  35. [43]

    S. Lei, X. Xiao, T. Zhang, H.-C. Li, Z. Shi, Q. Zhu, Exploring fine-grained image–text alignment for refer- ring remote sensing image segmentation, IEEE Transac- tions on Geoscience and Remote Sensing 63 (2025) 1–11. doi:10.1109/TGRS.2024.3522293

  36. [44]

    K. Li, G. V osselman, M. Y . Yang, Scale-wise bidirec- tional alignment network for referring remote sensing im- age segmentation, ISPRS Journal of Photogrammetry and Remote Sensing 226 (2025) 350–363. doi:10.1016/j. isprsjprs.2025.05.014

  37. [45]

    Zhang, Z

    T. Zhang, Z. Wen, B. Kong, K. Liu, Y . Zhang, P. Zhuang, J. Li, Referring remote sensing image segmentation via multi-scale spatially-guided joint prediction, IEEE Journal of Selected Topics in Applied Earth Observations and Re- mote Sensing (2025) 1–16doi:10.1109/JSTARS.2025. 3638802

  38. [46]

    Q. Ma, L. Li, X. Lu, L. Jiao, F. Liu, W. Ma, X. Liu, L. Sun, LSCF: Long-term semantic-guidance ConvFormer for re- ferring remote sensing image segmentation, IEEE Transac- tions on Geoscience and Remote Sensing 63 (2025) 1–13. doi:10.1109/TGRS.2025.3578515

  39. [47]

    Ho, D.-K

    C.-H. Ho, D.-K. Nguyen, M.-T. Pham, S. Lefèvre, RSSep: Sequence-to-sequence model for simultaneous referring remote sensing segmentation and captioning, in: Asian Conference on Computer Vision Workshops, 2024

  40. [48]

    Howard, M

    A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, M. Tan, W. Wang, Y . Zhu, R. Pang, V . Vasudevan, et al., Searching for MobileNetV3, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 1314–1324

  41. [49]

    J. Xu, Z. Xiong, S. P. Bhattacharyya, PIDNet: A real- time semantic segmentation network inspired by PID con- trollers, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 19529–19539

  42. [50]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, I. Polosukhin, Attention is all you need, Advances in Neural Information Processing Systems 30

  43. [51]

    S. Gao, Z. Chen, G. Chen, W. Wang, T. Lu, A VSegFormer: Audio-visual segmentation with transformer, in: Proceed- ings of the AAAI Conference on Artificial Intelligence, V ol. 38, 2024, pp. 12155–12163

  44. [52]

    J. Liu, Y . Wang, C. Ju, C. Ma, Y . Zhang, W. Xie, Annotation-free audio-visual segmentation, in: Proceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 5604–5614

  45. [53]

    Q. Yang, X. Nie, T. Li, P. Gao, Y . Guo, C. Zhen, P. Yan, S. Xiang, Cooperation does matter: Exploring multi-order bilateral relations for audio-visual segmentation, in: Pro- ceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, 2024, pp. 27134–27143

  46. [54]

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, A. Mohamed, HuBERT: Self-supervised speech representation learning by masked prediction of hidden units, IEEE/ACM Transactions on Audio, Speech, and Language Processing 29 (2021) 3451–3460

  47. [55]

    S. Chen, C. Wang, Z. Chen, Y . Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao, et al., WavLM: Large- scale self-supervised pre-training for full stack speech pro- cessing, IEEE Journal of Selected Topics in Signal Pro- cessing 16 (6) (2022) 1505–1518. 15

  48. [56]

    D. S. Park, W. Chan, Y . Zhang, C.-C. Chiu, B. Zoph, E. D. Cubuk, Q. V . Le, SpecAugment: A simple data augmen- tation method for automatic speech recognition, in: Inter- speech, 2019, pp. 2613–2617

  49. [57]

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, B. Guo, Swin Transformer: Hierarchical vision transformer using shifted windows, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 10012–10022. 16

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.