HorusEye: Language as Dynamic Attention for Emergency Visual Analysis

Armel Yara

REVIEW 4 major objections 6 minor 29 references

HorusEye: Language as Dynamic Attention for Emergency Visual Analysis

T0 review · 4 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read This paper claims that natural language feedback can serve as a dynamic attention mechanism for vision-language models in degraded emergency imagery, but only for some models and only in some conditions — Gemini gains 47.3% in thermal groun

desk verdict A useful benchmark, but the 'thermal' findings are grayscale and the feedback loop is oracle-assisted — the paper overclaims what it tests. read the letter →

arxiv 2606.14741 v2 pith:26ENZYIG submitted 2026-06-02 cs.CV cs.LG

Armel Yara This is my paper · ORCID

classification cs.CVcs.LG

keywords vision-languagemodelsvisualgroundinglanguagefeedbackdynamicattentionemergencyanalysisthermalimageryhallucinationRefCOCO-Degraded

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 tries to establish whether iterative natural-language prompts can redirect a vision-language model's attention to recover accuracy lost under fog, smoke, and thermal degradation. It builds RefCOCO-Degraded, a 15,244-image benchmark, and finds the answer is conditional: language feedback yields large gains for Gemini in thermal conditions but hurts Qwen2-VL under the same protocol. It also reports a Thermal Paradox, where cropping helps RGB classification but causes catastrophic accuracy drops in thermal imagery, and identifies BLIP-2 as dangerously prone to increased hallucination under degradation. If these results hold, emergency AI designers cannot assume prompt-based attention or cropping strategies transfer across models or conditions; they must validate each model-condition pair.

What carries the argument

The study's engine is RefCOCO-Degraded, a benchmark that takes 3,811 base images and applies four conditions: Clean, Fog, Smoke, and a Thermal proxy formed by grayscale luminance conversion. The central mechanism tested is iterative language feedback: when a first grounding prediction has IoU below 0.5, the model is told the error direction ('too far left', 'too high') and asked to correct its bounding box, up to three rounds. A second mechanism is explicit spatial attention via ground-truth cropping in posture classification. The paper also defines a Hallucination Score (H-Score) combining fabrication counts, overconfident language, and appropriate uncertainty.

What would settle it

Re-run the same four research questions on real thermal camera images (for example, from emergency response footage) instead of the synthetic luminance proxy. If real thermal images do not reproduce the +47.3% vs -5.1% feedback divergence, the Thermal Paradox, and BLIP-2's increased hallucination, then the central claims are artifacts of the chosen simulation.

Watch

Extended reading notes

Core claim

The central claim is that language feedback acts as a dynamic attention mechanism for some vision-language models but not others. Using a three-round iterative protocol applied only to samples where the initial grounding prediction failed (IoU below 0.5), Gemini improves by 47.3% in thermal conditions while Qwen2-VL degrades by 5.1% under the identical protocol. The paper also finds that cropping, a form of explicit spatial attention, is not universally beneficial: it improves RGB posture classification but reduces thermal accuracy by 26% for Gemini, because thermal images lose color and texture and therefore rely heavily on contextual cues. Finally, BLIP-2's hallucination score increases un

Load-bearing premise

The thermal condition is modeled as grayscale luminance conversion (0.299R + 0.587G + 0.114B), not as a real thermal sensor; if true thermal imagery behaves differently, every thermal-specific conclusion could collapse.

Editorial extensions

If this is right

For thermal emergency scenarios, Gemini appears to benefit substantially from iterative language feedback, while Qwen2-VL does not; deploying the same feedback protocol across models is unsafe.
Cropping should be avoided in thermal imagery, at least for Gemini, because it removes contextual cues needed for posture inference; full-image processing is the safer default.
BLIP-2's increasing hallucination under degradation, with no uncertainty hedging, makes it a poor candidate for emergency use unless mitigation is added.
The RefCOCO-Degraded benchmark provides a reusable protocol for testing other vision-language models and degradation types before deployment.
All conclusions are tied to the tested model versions; newer releases may behave differently.

Reading between the lines

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

The paper's thermal conclusions rest on a luminance-proxy simulation rather than real infrared sensor data; if real thermal cameras produce different visual statistics, the Thermal Paradox and the model-dependent feedback gap could change or disappear.
The model-dependence of language feedback suggests that a model's ability to follow directional language may correlate with its instruction-following strength or grounding architecture, but the paper does not isolate that mechanism.
LLaVA's constant 64% accuracy across all conditions and cropping states hints at a possible majority-class collapse; a direct test would be to check its per-class performance on a balanced posture dataset.
The H-Score is a surface-language heuristic; whether it predicts real operational harm, such as false triage decisions, is an untested extension.

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.

Desk Editor's Note

read the letter

Colleague,

The headline you need: the paper's central claim is not supported by its own protocol. The RefCOCO-Degraded benchmark is a real addition, and the empirical pattern of model-dependent responses to feedback is worth having, but the 'thermal' condition is grayscale and the feedback loop is oracle-driven. That makes the strong conclusions about emergency deployment premature.

What's good: the dataset construction is straightforward and reproducible — 3,811 base images with fog, smoke, and grayscale variants and clean controls. The observation that language feedback helps Gemini (+47.3% in the thermal/grayscale setting) while hurting Qwen2-VL (-5.1%) is interesting and, as far as I can tell, not already in the literature. The 'Thermal Paradox' — cropping helps on RGB but hurts on grayscale — is a useful caution even if the thermal label is wrong. The paper is clearly written and the four research questions are well structured.

Where it falls apart: the thermal condition is not thermal. It's the standard luminance equation (0.299R+0.587G+0.114B), so every thermal-specific conclusion is actually about grayscale. That could be fine if the paper said 'grayscale,' but it sells itself as emergency thermal analysis. RQ2's feedback is oracle-assisted: the prompt tells the model its error direction, which requires knowing the ground truth. That measures ability to follow a correction command, not ability to use language as an unprivileged dynamic attention signal. This is never acknowledged, and the Limitations section doesn't mention it. Additionally, RQ3 has only 50 samples per model, the RQ1 table omits BLIP-2 and LLaVA without explanation, and none of the tables report error bars. The H-Score is an ad hoc composite with no validation.

The paper is not incoherent on its own terms — the authors likely believe the protocol is fair — but the gap between what they claim and what the experiments support is large. The right fix is to reframe the paper as a study of synthetic fog/smoke/grayscale robustness, explicitly label the feedback results as oracle-assisted upper bounds, and add uncertainty quantification if the sample sizes allow.

My call: this deserves a serious referee because the benchmark and the phenomenon of model-dependent feedback are worth discussing, and the review process could steer the authors toward an honest reframing. But as submitted, it's a reject with clear instructions for major revision.

I'd bring it to a reading group as a case study in how easy it is to overclaim from synthetic degradation experiments.

Referee Report

4 major / 6 minor

Summary. The paper introduces RefCOCO-Degraded, a benchmark of 3,811 COCO images transformed into four conditions (Clean, Fog, Smoke, and Thermal, where 'thermal' is a luma grayscale conversion). It evaluates several VLMs across four research questions: visual grounding under degradation (RQ1), iterative language feedback as a dynamic attention mechanism (RQ2), posture-classification VQA with and without cropping (RQ3), and hallucination behavior measured by an author-defined H-Score (RQ4). The headline claims are that thermal imagery causes severe grounding degradation (e.g., Qwen2-VL -63.6% IoU), that language feedback is model-dependent (+47.3% for Gemini vs -5.1% for Qwen2-VL in thermal), that cropping produces a 'Thermal Paradox' (-26% for Gemini), and that BLIP-2 is unsafe because its H-Score increases under degradation. The paper concludes that language feedback can serve as dynamic attention for some models but not others, and that model-condition validation is required.

Significance. The proposed benchmark and the multi-model comparison target a relevant and under-studied problem: VLM reliability in emergency conditions. The paper is transparent about some limitations, and the degradation pipeline is systematic. If the central findings were valid, they would offer useful evidence that language feedback is model- and condition-dependent, and the 'Thermal Paradox' could inform deployment choices. However, the two load-bearing pillars of the work—the thermal proxy and the oracle-assisted RQ2 protocol—undermine the central claims as stated. The paper provides no code, data, or statistical inference, and the H-Score is an unvalidated metric. The potential significance is real, but the current evidence does not establish it.

major comments (4)

[§5.2.1, RQ2 method] The R2 feedback prompt is constructed from the known difference between the R1 prediction and the ground-truth box ('Your previous prediction was [too far left/right/too high/low]'). This is oracle feedback: in a real emergency deployment, the error direction is unknown because the ground truth is not available. Therefore the central model-dependence result—Gemini +47.3% vs Qwen2-VL -5.1% in thermal—measures the models' ability to follow perfectly correct directional correction commands, not their ability to use language as an unprivileged dynamic-attention signal. The paper never labels RQ2 as oracle-assisted, and §8 does not mention this limitation. The main claim is not supported by the current protocol; a non-oracle feedback condition (e.g., 'your previous box was wrong; re-estimate the location' without giving the direction) or an explicit framing as an upper-bound study is needed.
[§4.2, thermal condition definition] The 'Thermal' condition is defined as I_thermal = 0.299R + 0.587G + 0.114B, i.e., a standard luma grayscale conversion. Real thermal imagery encodes emitted long-wave infrared radiation, not visible-light luminance; contrast, noise, thermal signatures, and object appearances are fundamentally different. All thermal-specific conclusions—the RQ1 collapse, the RQ2 +47.3% recovery, the Thermal Paradox (-26%), and the BLIP-2 hallucination increase—depend entirely on this grayscale proxy. The limitations sentence in §8 that synthetic degradation 'might not perfectly align' understates the severity: the experiments cannot be described as thermal without validation on real thermal sensor data, or the claims must be reframed as 'luminance/grayscale degradation.'
[§5.3/Table 5 and §8] RQ3 uses only 50 annotated samples per model, as acknowledged in §8. For n=50, the standard error of an accuracy estimate is at most about 7 percentage points (at 50% accuracy), so several reported differences—Qwen2-VL Clean -12%, Qwen2-VL Thermal +2%, LLaVA 0%—are within typical sampling noise. No confidence intervals or significance tests are given. The 'Thermal Paradox' (-26% for Gemini, from 94% to 68%) has a 95% CI width of roughly ±12 percentage points at this sample size, making the claimed 'catastrophic failure' borderline rather than definitive. The paper must either substantially increase the sample size, report uncertainty, and provide inferential statistics, or soften the 'catastrophic' language accordingly.
[§5.4/Table 6, H-Score] The H-Score is introduced as H = Fabrication + 0.5×Overconfidence - 0.3×Uncertainty, with weights that are justified neither theoretically nor empirically, and 'Fabrication' is detected via unspecified keyword matching. There is no validation, no inter-annotator agreement measure, and no comparison against established hallucination metrics such as CHAIR or POPE. The safety-relevant conclusion that BLIP-2 is 'dangerous' and 'unsuitable for emergency deployment' rests entirely on this unvalidated composite. The paper should validate the metric against existing benchmarks, report the component counts separately, present qualitative examples, or avoid strong safety claims until this is done.

minor comments (6)

[Abstract/Table 3] The abstract and Section 3.2 state that five VLMs are evaluated, but Table 3 (RQ1) reports results only for Gemini, Qwen2-VL, and Kosmos-2. BLIP-2 and LLaVA are missing from the main grounding benchmark. Please clarify the per-RQ model coverage or add the missing rows.
[§5.1.2] The prompt template, decoding parameters, temperature, and exact model versions/checkpoints are not provided. This is essential for reproducibility, especially for proprietary models like Gemini where the version name is the only control.
[General] No confidence intervals, standard deviations, or significance tests are reported in any table. The paper should include these throughout, especially for the small-sample RQ3 and RQ2 gain percentages.
[§4.2/§5.3] Equations are not numbered, which makes referencing difficult. Also, Figure 2 (coverage matrix) and Figure 5 (pipeline) are not described in the text; please add in-text references and captions that make the figures self-contained.
[§3.3] The H-Score formula uses an en-dash in the table but a minus sign in the text; please standardize the notation. Also, the description 'Higher H-Score = more dangerous' conflicts with the negative values in Table 6; clarify the intended interpretation.
[References] References [18] and [19] appear to be dated after the other citations; please verify the publication years and ensure all URLs/arXiv IDs are correct.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: all central claims are empirical measurements, not derivations that reduce to their inputs.

full rationale

I walked the paper's claimed derivation chain. The paper is an empirical benchmark study: it constructs RefCOCO-Degraded from RefCOCO plus synthetic degradations, measures IoU/accuracy/H-Score across models, and reports differences. No parameter is fitted to a subset of data and then renamed a prediction; the H-Score is an author-defined scoring rubric, not a fitted predictor of the results that confirm it. The RQ2 feedback prompt is indeed oracle-assisted (the correction direction is derived from ground-truth error: 'Your previous prediction was [too far left/right/too high/low],' Sec. 5.2.1), which limits real-world deployment validity and should have been stated as a limitation. However, that is an experimental-design/construct-validity threat, not circularity: the reported IoU gains are not forced by construction, and the paper does not claim to predict the error direction from the model's own outputs. There are no load-bearing self-citations, no imported uniqueness theorems, and no ansatz smuggled in via citation. The thermal proxy (grayscale luminance) is a modeling assumption, not a circular definition of the findings. Therefore the core claims retain independent empirical content, and no circular step can be exhibited with the paper's own equations.

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

The central claims pull in several hand-chosen quantities and unvalidated domain assumptions: the 'thermal' proxy, the oracle feedback signal, the H-Score weights, and the fixed degradation intensity. None of these are externally grounded, which raises correctness risk even though the paper is an empirical study rather than a derivation.

free parameters (4)

Degradation intensity = 0.5 (fog, smoke, thermal)
All degradation conditions use intensity 0.5; results may depend on this arbitrary severity level, and no sensitivity analysis is reported.
H-Score weights = 1.0, 0.5, -0.3 (Fabrication, Overconfidence, Uncertainty)
The composite hallucination score weights are chosen by hand and not calibrated against existing hallucination benchmarks.
RQ2 failure filter threshold = IoU < 0.5
Samples are included in the feedback study only when R1 IoU < 0.5; the threshold changes which samples each model/condition contributes, and no sensitivity analysis is given.
RQ3 sample count = 50 per model
Acknowledged in Limitations; posture accuracy differences of a few percentage points rest on ~50 samples, so no error bars can be meaningful.

assumptions (4)

ad hoc to paper A luma grayscale conversion simulates thermal camera imagery
Sec. 4.2 defines I_thermal = 0.299R+0.587G+0.114B. All thermal-specific conclusions assume this proxy behaves like a thermal sensor, which is not established: real thermal images reflect emitted infrared, not color luminance.
domain assumption Ground-truth error direction is available for language feedback
Sec. 5.2.1 says R2 feedback is based on R1 error direction ('too far left/right/too high/low'). This requires a supervisor who already knows the correct box, which is not available in autonomous emergency deployment.
domain assumption RefCOCO person-referring expressions support health posture classification
RQ3 transfers referring-expression grounding to posture triage with 50 annotated samples, assuming the task is representative of emergency health assessment.
ad hoc to paper Keyword counting captures hallucination severity
Sec. 5.4.1 defines H-Score via 'known fabrication indicators' keyword matching without specifying the list or validating against CHAIR/POPE.

invented entities (2)

Thermal Paradox
purpose: Label for the observation that cropping helps RGB posture classification but hurts the simulated thermal condition
No evidence from real thermal sensors; under the grayscale proxy, the phenomenon may simply be 'cropping hurts grayscale classification', which is not a paradox.
H-Score
purpose: Composite hallucination severity measure used to rank models as SAFER/DANGER
Uncalibrated; keyword matching lists and weights are not published and it is not benchmarked against CHAIR/POPE.

how reviews work

0 comments

Cite this review

Pith. "Pith review of HorusEye: Language as Dynamic Attention for Emergency Visual Analysis." pith.science (2026). https://pith.science/paper/26ENZYIG

@misc{pith2026260614741,
  author       = {Pith},
  title        = {Pith review of: HorusEye: Language as Dynamic Attention for Emergency Visual Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/26ENZYIG}},
  note         = {Machine review of arXiv:2606.14741}
}

read the original abstract

We introduce HorusEye, Language as Dynamic Attention for Emergency Visual Analysis. Our investigation followed five stages. The first one is benchmarking RefCOCO-Degraded, a dataset of 15,244 images (3,811 base images x 4 conditions: Clean, Fog, Smoke and Thermal) with systematic visual degradation. Through four research questions, we evaluate multiple VLMs (Gemini, Qwen2-VL, BLIP-2, LLaVA, Kosmos-2) across visual grounding the second stage, language feedback recovery the third one, health VQA tasks the fourth, and hallucination analysis the final stage. Our key finding is that language feedback effectiveness is model-dependent: Gemini achieves +47.3% improvement in thermal conditions through iterative language feedback, while Qwen2-VL shows -5.1% degradation under the same protocol. We also identify the 'Thermal Paradox' where cropping strategies that improve RGB performance catastrophically fail in thermal imagery. Furthermore, BLIP-2 uniquely hallucinates more under degradation, making it unsuitable for emergency deployment

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 7 linked inside Pith

[1]

Xu, K., Ba, J., Kiros, R., Cho, K., Courville, A., Salakhudinov, R., Zemel, R., & Bengio, Y. (2015). Show, Attend and Tell: Neural Image Caption Generation with Visual Attention. In ICML

2015
[2]

Anderson, P., He, X., Buehler, C., Teney, D., Johnson, M., Gould, S., & Zhang, L. (2018). Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering. In CVPR

2018
[3]

Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014). Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation. In CVPR

2014
[4]

Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., & Houlsby, N. (2021). An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. In ICLR . arXiv:2010.11929

arXiv 2021
[5]

N., Kaiser, Ł., & Polosukhin, I

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention Is All You Need. In NeurIPS

2017
[6]

C., & Berg, T

Yu, L., Poirson, P., Yang, S., Berg, A. C., & Berg, T. L. (2016). Modeling Context in Referring Expressions. In ECCV

2016
[7]

Hendrycks, D., & Dietterich, T. (2019). Benchmarking Neural Network Robustness to Common Corruptions and Perturbations. In ICLR

2019
[8]

Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014). Microsoft COCO: Common Objects in Context. In ECCV

2014

Show all 29 references

[9]

Kazemzadeh, S., Ordonez, V., Matten, M., & Berg, T. L. (2014). ReferItGame: Referring to Objects in Photographs of Natural Scenes. In EMNLP

2014
[10]

Hendrycks, D., Basart, S., Mu, N., Kadavath, S., Wang, F., Dorber, E., Desai, R., Zhu, T., Parajuli, S., Guo, M., Song, D., Steinhardt, J., & Gilmer, J. (2021). The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization. In ICCV

2021
[11]

Li, J., Li, D., Savarese, S., & Hoi, S. (2023). BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models. In ICML

2023
[12]

Liu, H., Li, C., Wu, Q., & Lee, Y. J. (2023). Visual Instruction Tuning (LLaVA). In NeurIPS . [13] Gemini Team, Google. (2024). Gemini: A Family of Highly Capable Multimodal Models. arXiv:2312.11805

2023 arXiv
[14]

Gemini Team, Google. (2024). Gemini 1.5: Unlocking Multimodal Understanding Across Millions of Tokens of Context. arXiv:2403.05530

2024 arXiv
[15]

Wang, P., Bai, S., Tan, S., Wang, S., Fan, Z., Bai, J., Chen, K., Liu, X., Wang, J., Ge, W., Fan, Y., Dang, K., Du, M., Ren, X., Men, R., Liu, D., Zhou, C., Zhou, J., & Lin, J. (2024). Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution. arXiv...

2024 arXiv
[16]

Peng, Z., Wang, W., Dong, L., Hao, Y., Huang, S., Ma, S., & Wei, F. (2023). Kosmos-2: Grounding Multimodal Large Language Models to the World. In ICLR 2024 . arXiv:2306.14824

2023 arXiv
[17]

Liu, H., Li, C., Li, Y., & Lee, Y. J. (2024). Improved Baselines with Visual Instruction Tuning (LLaVA-1.5). In CVPR

2024
[18]

Saxena, R., Suglia, A., & Minervini, P. (2025). VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models. arXiv:2603.06148

2025
[19]

Lin, Z., Wan, H., Zhong, Y., & Ji, X. (2025). DIQ-H: Evaluating Hallucination Persistence in VLMs Under Temporal Visual Degradation. arXiv:2512.03992

2025 arXiv
[20]

A., Burns, K., Darrell, T., & Schiele, B

Rohrbach, A., Hendricks, L. A., Burns, K., Darrell, T., & Schiele, B. (2018). Object Hallucination in Image Captioning (CHAIR). In EMNLP

2018
[21]

X., & Wen, J.-R

Li, Y., Du, Y., Zhou, K., Wang, J., Zhao, W. X., & Wen, J.-R. (2023). Evaluating Object Hallucination in Large Vision-Language Models (POPE). In EMNLP

2023
[22]

Kaul, P., Xie, Z., & Zisserman, A. (2024). THRONE: An Object-based Hallucination Benchmark for the Free-form Generations of Large Vision-Language Models. In CVPR

2024
[23]

Mao, J., Huang, J., Toshev, A., Camburu, O., Yuille, A., & Murphy, K. (2016). Generation and Comprehension of Unambiguous Object Descriptions. In CVPR

2016
[24]

Deng, J., Yang, Z., Chen, T., Zhou, W., & Li, H. (2021). TransVG: End-to-End Visual Grounding with Transformers. In ICCV

2021
[25]

Sakaridis, C., Dai, D., & Van Gool, L. (2018). Semantic Foggy Scene Understanding with Synthetic Data. International Journal of Computer Vision , 126(9), 973–992

2018
[26]

Li, B., Ren, W., Fu, D., Tao, D., Feng, D., Zeng, W., & Wang, Z. (2019). Benchmarking Single-Image Dehazing and Beyond. IEEE Transactions on Image Processing , 28(1), 492–505

2019
[27]

L., & Parikh, D

Antol, S., Agrawal, A., Lu, J., Mitchell, M., Batra, D., Zitnick, C. L., & Parikh, D. (2015). VQA: Visual Question Answering. In ICCV

2015
[28]

Goyal, Y., Khot, T., Summers-Stay, D., Batra, D., & Parikh, D. (2017). Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering. In CVPR

2017
[29]

Murphy, R. R. (2014). Disaster Robotics. MIT Press. [30] Radovic, M., Adarkwa, O., & Wang, Q. (2017). Object Recognition in Aerial Images Using Convolutional Neural Networks. Journal of Imaging , 3(2), 21

2014
[31]

Dehghani, M., Mustafa, B., Djolonga, J., Heek, J., Minderer, M., Caron, M., Steiner, A., Puigcerver, J., Geirhos, R., Alabdulmohsin, I., et al. (2024). Patch n' Pack: NaViT, a Vision Transformer for any Aspect Ratio and Resolution. In NeurIPS . arXiv:2307.06304

2024 arXiv

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

Xu, K., Ba, J., Kiros, R., Cho, K., Courville, A., Salakhudinov, R., Zemel, R., & Bengio, Y. (2015). Show, Attend and Tell: Neural Image Caption Generation with Visual Attention. In ICML

2015

[2] [2]

Anderson, P., He, X., Buehler, C., Teney, D., Johnson, M., Gould, S., & Zhang, L. (2018). Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering. In CVPR

2018

[3] [3]

Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014). Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation. In CVPR

2014

[4] [4]

Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., & Houlsby, N. (2021). An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. In ICLR . arXiv:2010.11929

arXiv 2021

[5] [5]

N., Kaiser, Ł., & Polosukhin, I

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention Is All You Need. In NeurIPS

2017

[6] [6]

C., & Berg, T

Yu, L., Poirson, P., Yang, S., Berg, A. C., & Berg, T. L. (2016). Modeling Context in Referring Expressions. In ECCV

2016

[7] [7]

Hendrycks, D., & Dietterich, T. (2019). Benchmarking Neural Network Robustness to Common Corruptions and Perturbations. In ICLR

2019

[8] [8]

Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014). Microsoft COCO: Common Objects in Context. In ECCV

2014