Pith. sign in

REVIEW 4 major objections 4 minor 52 references

Counting exposed buildings from pre-strike maps and text-derived blast radii can bypass satellite embargoes, and depth-augmented vision-language models beat 2D segmentation in dense cities.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 01:14 UTC pith:FV43CWC3

load-bearing objection A useful LVLM counting benchmark wrapped in an untested damage-estimation claim — the internal comparison is solid, but the ground truth is defined by the same blast-radius parameters the pipeline uses, so the benchmark measures circle-counting skill, not war damage. the 4 major comments →

arxiv 2608.00119 v1 pith:FV43CWC3 submitted 2026-07-31 cs.CV cs.AI

Counting the Cost of War Under Satellite Embargo: Zero-Shot Estimation of Impacted Infrastructure

Saleh Sakib Ahmed , M. Sohel Rahman This is my paper
classification cs.CV cs.AI
keywords zero-shot countingbuilding exposure estimationlarge vision-language modelsdepth mapsHopkinson-Cranz scalingsatellite embargoconflict mappingremote sensing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper reframes war-damage estimation as a zero-shot geometric projection task on archival pre-strike imagery, allowing estimates to run during satellite embargoes. Reported strike coordinates retrieve old satellite maps; a language model estimates weapon payload W, and the Hopkinson-Cranz cube-root law R_base = Z·W^(1/3) draws the blast perimeter. The open question—how many buildings stand inside that perimeter—is answered either by 2D segmentation with an adaptive field-of-view or by large vision-language models given segmentation masks plus 2.5D pseudo-height depth overlays. The core empirical claim is that depth-augmented LVLMs resolve overlapping rooftops and outperform segmentation in congested urban cores, while cheap segmentation remains best in sparse scenes, together forming a density-routed zero-shot crisis-mapping pipeline. A sympathetic reader would care because it promises immediate structural-exposure estimates when post-strike imagery is commercially embargoed, and it sets a benchmark for spatial reasoning in vision-language models.

Core claim

The paper's central claim is that 2.5D pseudo-height depth maps—generated by applying monocular depth estimation to top-down satellite rasters—act as a spatial regularizer that lets open-weight large vision-language models visually decouple touching rooftops, eliminating the boundary-collapse failure of flat 2D segmentation in dense urban cores. In the reported benchmark, the best depth-augmented configuration reduces mean absolute error by roughly 40–50% relative to the unassisted vision-language baseline and produces lower error than the deterministic segmentation reference in high- and extreme-density scenes, while the adaptive field-of-view routine removes zoom bias in the segmentation s

What carries the argument

The Hopkinson-Cranz cube-root scaling law, R_base = Z·W^(1/3), with Z = 5 m/kg^(1/3) as a far-field reference distance, converts an LLM-extracted munition class and estimated TNT-equivalent mass W into a deterministic blast radius R_moderate after a severity modifier; this turns damage assessment into a geometric counting problem on pre-strike tiles. To carry out the counting, an adaptive field-of-view routine selects the best zoom level for a foundational 2D segmentation model, and a 2.5D pseudo-height depth overlay—monocular depth maps rendered as colormapped rasters—provides the relative elevation cues that let vision-language models separate merged rooftops. A median-of-five inference pr

Load-bearing premise

The load-bearing premise is that the radius computed from an LLM-extracted payload and the Hopkinson-Cranz coefficient marks the right boundary for 'impacted infrastructure'—human ground truth is annotated inside that same computed radius, so a wrong radius moves both the target and the reference together, and the pipeline could score well while predicting the wrong buildings.

What would settle it

After a satellite embargo lifts, collect post-strike imagery for a sample of the 890 benchmark events and compare the pipeline's exposure counts against manually mapped damaged structures. If damaged-building counts are only weakly correlated with exposure counts, or if the optimized radius does not separate damaged from undamaged buildings, the central claim fails. A cheaper test: fix the same scenes, change only the payload table or Z value, and re-annotate ground truth inside the new radius; the paper's claim of invariance would predict the comparative ranking of counting systems is preserv

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • During satellite data embargoes, humanitarian and GEOINT teams can obtain structural-exposure estimates within minutes of a text alert, using only archival imagery; no post-strike satellite access is required.
  • A density-routed hybrid deployment is the efficient operating mode: lightweight 2D segmentation for sparse rural scenes, depth-augmented vision-language models for dense urban footprints (16+ structures), where the depth overlay cuts extreme-density error variance by roughly 40-60%.
  • The pipeline yields a reproducible zero-shot benchmark for spatial reasoning in vision-language models: count only buildings whose roofs intersect a circular boundary while ignoring visually adjacent blocks outside it.
  • Open-source incident feeds can be quality-filtered: the paper's drift analysis suggests treating crowdsourced coordinates with more than ~250 m of drift as unreliable, since beyond that point building-count errors grow sharply.
  • The blast-radius parameters (payload table, scaling coefficient, severity modifier) are user-calibratable knobs; the paper argues the comparative ranking of counting architectures is invariant to boundary recalibration.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The benchmark never tests the blast-radius parameters against actual damage: human annotators drew ground truth inside the very radius computed by R = Z·W^(1/3) with the severity modifier, so if Z, W, or the modifier are wrong, the target and the ground truth shift together; the experiments therefore measure circle-counting skill, not war-damage prediction. The paper itself acknowledges this in §3
  • The counts are structural exposure—standing buildings inside a predicted perimeter—not verified destruction, as the paper states in §2.4.3; using the pipeline as a damage census would overstate destruction where interceptions or strikes on unpopulated terrain occur.
  • The depth-overlay trick likely transfers to other overhead counting tasks where flat imagery merges objects (e.g., vehicles, refugee camp shelters, construction sites) and could be tested immediately on public satellite tiles.
  • The 250 m drift threshold could be built into a live ingestion system: combining crowdsourced and curated feeds, discarding high-drift events, and routing scenes by density would make the pipeline operational rather than a benchmark.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes an end-to-end pipeline for zero-shot estimation of building exposure (and, by title, "impacted infrastructure") from pre-strike satellite imagery. Text alerts from LiveUAMap and ArcGIS are parsed by an LLM to extract weapon class and severity; a Hopkinson–Cranz scaling law R = Z·W^(1/3) is then used to define a blast perimeter, and buildings within that perimeter are counted either by SAMGeo segmentation with an adaptive field-of-view or by LVLMs augmented with segmentation masks and monocular depth maps. The authors benchmark four LVLM input configurations on two 2026 Middle East datasets, report that depth augmentation improves counting accuracy, and propose a density-stratified deployment rule. The internal counting comparison is reasonably careful: N=5 median filtering, Wilcoxon signed-rank tests, human consensus annotation, and density-stratified analysis are all present.

Significance. If the claims were fully supported, the paper would make a useful contribution to crisis mapping under satellite embargoes and to LVLM spatial-reasoning benchmarking. The strengths of the internal evaluation should be acknowledged: the use of multiple independent inference runs with median aggregation, non-parametric significance testing, stratified error analysis by density tier, and a human-consensus annotation protocol are all appropriate for comparing counting models inside a fixed perimeter. The paper also ships code and data links, which aids reproducibility. However, the load-bearing operational claim — that the system estimates "impacted infrastructure" — is not validated against any independent post-strike damage data. The human-annotated ground truth is itself defined inside the model-generated blast radius, so the benchmark measures the ability to count buildings inside a parameterized circle, not the physical accuracy of the radius projection. The paper partially acknowledges this in Sections 2.4.3 and 3.8, but the abstract, title, and deployment recommendations (Table 3, Panel B) are framed much more strongly. The result is a credible comparison of counting architectur

major comments (4)
  1. [§4.2.4, §3.8, Abstract] The ground-truth counts are produced by asking annotators to count buildings inside the same R_moderate computed by Eqs. (2)–(3) using the authors' Z, W lookup table, and Δ_modifier. Any error in these parameters changes the target circle and the annotated reference count together, so the reported MAE/MSE cannot validate the blast-radius projection. The paper concedes this in §3.8 ('ground truth within this benchmark is best understood as a parameterized reference baseline') and §2.4.3 ('theoretical exposure, not verified physical destruction'), but the title, abstract, and Table 3 Panel B use the stronger 'impacted infrastructure' framing. Either validate against independent post-strike damage data (even a small set) or reframe all claims and deployment recommendations as structural-exposure estimation within a parameterized perimeter.
  2. [§2.4.1, §2.4.2, §4.1.1] The paper reports that LiveUAMap coordinates have a median drift of 4.07 km and a mean of 34.87 km relative to ArcGIS, and then proposes a 250 m quality-filtering threshold in §2.4.2. However, the methodology and evaluation sections never state whether this threshold was actually applied to the LiveUAMap test set before image retrieval and annotation. With blast radii of 5–60 m, km-scale coordinate errors make the projected circle physically meaningless. Please state explicitly whether the LiveUAMap benchmark was filtered, how many events survived, and how the drift analysis relates to the reported MAE/MSE results. Without this, the crowdsourced stream's results do not support the operational claim.
  3. [§2.2, Table 3 Panel A] The abstract claims that 'depth-augmented LVLMs dramatically outperform traditional segmentation in congested urban centers,' but the statistical results in Table 3 Panel A show that significant depth gains are limited to specific architectures. Qwen 3.6 on LiveUAMap shows no significant Baseline→+Depth improvement (p=0.3198), and Nvidia Cosmos and GLM 4.6V fail to benefit from spatial augmentations. Only Gemma 4 (and, on some comparisons, Claude) consistently show large gains. Please restrict the claim to the architectures that actually benefit and present the density-stratified results per architecture rather than as an aggregated LVLM cohort.
  4. [§3.7, Eq. (1)] The external sanity check in §3.7 is not a validation. Applying a uniform 40-units-per-building multiplier (10 floors × 4 units/floor) to the national aggregate counts and comparing with an 81,000-unit figure from a state social-media source assumes a single national density profile and an unverified ground-truth number. This cannot substitute for per-event damage validation, and the text should not present it as evidence that the structural counts 'align realistically' with actual damage.
minor comments (4)
  1. [§4.2.2 vs §2.2, Fig. 1] Configuration A/B definitions are reversed: in §2.2 and Fig. 1, Config A is location-labeled imagery and Config B is unlabeled raw imagery; in §4.2.2, Config A is raw and Config B is labeled. Please harmonize these labels to avoid ambiguity when reading Table 2.
  2. [§2.1 and throughout] Typo: 'spatial contect' should be 'spatial context'. Also 'GeoSAM' and 'SAMGeo' are used interchangeably; please use one name consistently.
  3. [Tables 2, S16] The +Depth notation is defined as segmentation-plus-depth, but the column header is easy to misread as depth alone. Consider using '+Seg+Depth' or a footnote in every table.
  4. [References] Reference [26] is a social-media post; for a quantitative validation claim, please use a verified secondary source or clearly label the figure as a preliminary official estimate.

Circularity Check

2 steps flagged

The 'impacted infrastructure' ground truth is defined inside the same blast radius the pipeline computes, so the exposure claim reduces to circle-counting by construction; only the model-to-model comparison is independent.

specific steps
  1. self definitional [Sec. 4.2.4 Ground Truth Annotation (cf. Sec. 4.1.3, Eqs. 2–3)]
    "For each benchmark location, the target radius was first computed using the blast-radius estimation procedure described in Sec. 4.1.3, defining the explicit spatial boundaries for the structural counting task. Human annotators then established the reference building counts within these zones using a collaborative panel of five independent researchers."

    The target quantity 'impacted structures' is defined as buildings inside R_moderate = floor(Z·W^(1/3)·max(0.2, 1+Δ)), and the human reference counts are annotated inside exactly that same radius computed by the paper's own procedure. Thus the benchmark never compares the predicted blast perimeter to any independent measure of real damage: if Z, W, or Δ are wrong, the evaluation circle rescales and the ground-truth annotation circle rescales with it. Low MAE therefore validates counting skill inside the model's own circle, not the physical exposure estimate.

  2. self definitional [Sec. 3.8 Limitations and Future Work]
    "Because the structural reference counts are inherently defined by these spatial bounds, any adjustment to the baseline payload parameters naturally scales the evaluation area, which may slightly alter the resulting damage count. For this reason, the ground truth within this benchmark is best understood as a parameterized reference baseline calibrated to specific physical inputs, rather than a static absolute value."

    This is the paper's own admission that the ground-truth 'damage count' is a function of the same blast-radius parameters the pipeline uses. Errors in the payload table or the severity heuristic move the prediction target and the annotation target together, so the reported MAE/MSE cannot validate the radius model or the 'impacted infrastructure' claim. The internal model comparisons remain meaningful for counting inside a fixed circle, but the paper's operational framing as zero-shot estimation of impacted infrastructure is untested and reduces by construction to counting inside its own projected perimeter.

full rationale

The paper contains a real internal-validity circularity, but it is localized. The counting benchmark itself is not circular for comparing LVLMs against SAMGeo: all models see the same target circle and the same human-annotated building counts, so the finding that depth-augmented Gemma outperforms segmentation in dense scenes is a legitimate comparison of counting capability. However, the paper's central operational claim—'zero-shot estimation of impacted infrastructure' / 'counting the cost of war'—is evaluated against ground truth constructed from the same blast-radius formula the system itself uses. Section 4.2.4 states that the target radius was 'first computed using the blast-radius estimation procedure described in Sec. 4.1.3' and that human counts were then made 'within these zones'; Section 3.8 concedes that ground truth is 'a parameterized reference baseline calibrated to specific physical inputs' whose counts change whenever the payload parameters are adjusted. Section 2.4.3 further concedes these counts are 'theoretical exposure, not verified physical destruction.' Consequently, Z, W, and Δ_modifier are never validated against actual blast outcomes—they define both the prediction and the truth. Any systematic error in the radius simply rescales the scoring circle, and a model can achieve low MAE while the projected damage perimeter is physically wrong by a large factor. This is a self-definitional circularity for the exposure-estimation claim, meriting a score of 6. There is no load-bearing self-citation or uniqueness-theorem issue in the paper.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The central estimate depends on blast-radius parameters Z and W chosen by hand rather than derived from data or validated against actual damage. The evaluation boundary and the human-annotated ground truth are both defined by that radius, so the benchmark does not independently validate the impact model. The counting-model comparison itself is grounded in human annotation and statistical tests.

free parameters (5)
  • Scaled distance Z = 5 m/kg^(1/3)
    Hand-selected to lie in the far-field regime (>4 m/kg^(1/3)); directly sets every blast radius via R = Z·W^(1/3). Not derived from damage data.
  • Representative TNT-equivalent masses W per weapon class = 1, 14, 27, 64, 343, 512, 1728 kg
    Class-level values chosen from ranges in cited sources; e.g., 64 kg for heavy artillery despite 23 kg for GMLRS and 90 kg for Fajr-5. Changing W redefines the evaluation boundary and therefore the ground-truth counts.
  • Severity modifier Δ_modifier = {-0.5, 0, +0.5}
    Heuristic scaling of R_base by max(0.2, 1.0 + Δ_modifier); assigned by an LLM from text, not calibrated against observed damage.
  • SAMGeo inclusion weights = 1.0 full / 0.5 partial / 0.0 outside
    Manual scoring weights used to count partially and fully intersecting buildings in Stream 1; affects the segmentation baseline counts.
  • NMS IoU threshold and nesting threshold = IoU ≥ 0.5; nesting > 30%
    Chosen thresholds for duplicate suppression and nested-shape removal; affect the final mask set and thus all downstream counts.
axioms (4)
  • domain assumption Hopkinson-Cranz cube-root scaling law gives a meaningful damage/exposure boundary for the weapon classes considered
    The entire radius estimate rests on this scaling law and on the assumption that the selected scaled distance Z=5 represents structural exposure. Invoked in Sec. 4.1.3 and Supp. S2.
  • ad hoc to paper Counting buildings inside the model-generated radius on pre-strike imagery is a valid proxy for impacted infrastructure
    Human ground truth is annotated inside that same radius (Sec. 4.2.4), and the paper itself concedes counts are 'theoretical exposure, not verified physical destruction' (Sec. 2.4.3).
  • domain assumption Monocular depth maps applied to top-down satellite imagery encode meaningful relative building-height cues
    Depth Anything-type models are trained mostly on ground-level imagery; the paper assumes they transfer to nadir/oblique satellite rasters. Cited in Sec. 1 and used in Config D.
  • domain assumption ArcGIS StoryMaps exact coordinates are ground-truth incident locations
    Used to pair LiveUAMap events and to define the test set; the paper does not independently verify these coordinates against physical evidence.

pith-pipeline@v1.3.0-alltime-deepseek · 39207 in / 14209 out tokens · 162950 ms · 2026-08-04T01:14:30.652251+00:00 · methodology

0 comments
read the original abstract

Rapid estimation of impacted structures - critical for conflict-zone humanitarian response - is frequently hindered by post-strike satellite data embargoes and imagery blackouts. We bypass this operational bottleneck by reframing impacted building mapping as a zero-shot geometric projection task on archival, pre-strike maps. Using coordinate and incident text from LiveUAMap and ArcGIS, Large Language Models extract weapon payloads (W) to project kinetic blast perimeters via Hopkinson-Cranz scaling (R_base = Z * W^(1/3)). To count exposed structures within these zones without post-strike imagery, we introduce two technical innovations: Adaptive Field-of-View to eliminate resolution (zoom) bias in 2D segmentation (SAMGeo), and 2.5D pseudo-height depth maps combined with segmentation masks to help Large Vision-Language Models (LVLMs) resolve overlapping, dense rooftops. Evaluated on 2026 Middle East conflict data, depth-augmented LVLMs dramatically outperform traditional segmentation in congested urban centers. This establishes a powerful hybrid paradigm for zero-shot crisis mapping: ultra-fast 2D segmentation for sparse rural zones, and depth-augmented LVLMs for dense urban environments.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

52 extracted references · 4 linked inside Pith

  1. [1]

    Accessed: 2026-07-23 (2026)

    Al Jazeera: US satellite firm Planet Labs announces blackout on war on Iran images. Accessed: 2026-07-23 (2026). https://www.aljazeera.com/news/2026/4/5/ us-satellite-firm-planet-labs-announces-blackout-on-war-on-iran-images

    2026

  2. [2]

    Accessed: 2026-07-26 (2020)

    European Space Agency: What is Earth Observation? https://www.esa.int/Applications/Observing the Earth/ What is Earth observation. Accessed: 2026-07-26 (2020)

    2026

  3. [3]

    https://business.esa.int/ newcomers-earth-observation-guide

    European Space Agency: Newcomers Earth Observation Guide. https://business.esa.int/ newcomers-earth-observation-guide. Accessed: 2026-07-26 (2023)

    2026

  4. [4]

    Geological Survey: New Landsat 9 Cloud Cover Assessment Validation Dataset Published

    U.S. Geological Survey: New Landsat 9 Cloud Cover Assessment Validation Dataset Published. https://www. usgs.gov/landsat-missions/news/new-landsat-9-cloud-cover-assessment-validation-dataset-published. Accessed: 2026-07-26 (2023)

    2026

  5. [5]

    Space Force, Space Systems Command: Focused on the Threat: Directed Energy Weapons

    U.S. Space Force, Space Systems Command: Focused on the Threat: Directed Energy Weapons. https://www.ssc.spaceforce.mil/Newsroom/Article-Display/Article/3913339/ focused-on-the-threat-directed-energy-weapons-part-3-of-6. Accessed: 2026-07-26 (2024)

    arXiv 2026

  6. [6]

    Space Force, Space Systems Command: Space Threats

    U.S. Space Force, Space Systems Command: Space Threats. https://www.ssc.spaceforce.mil/Space-Threats. Accessed: 2026-07-26 (2024)

    2026

  7. [7]

    https://iran.liveuamap.com/

    Liveuamap: Iran News on Live Map – War in Iran – Conflict in the Gulf. https://iran.liveuamap.com/. Accessed: 2026-05-14 (2026)

    2026

  8. [8]

    and Israeli Strikes in Iran (2026)

    Mealie, D., Hurwitz, H., Cordola, B., Schulert, D., Ewert, M., Kramer, N., Corticchiato, L., Weinstein, C., Ornelas, E.: Interactive Map: U.S. and Israeli Strikes in Iran (2026). ArcGIS StoryMaps, Institute for the Study of War (ISW) & AEI’s Critical Threats Project (CTP). Accessed: August 4, 2026(2026). https://storymaps. arcgis.com/stories/089bc1a2fe684...

    2026

  9. [9]

    Journal of Open Source Software8(89), 5663 (2023)

    Wu, Q., Osco, L.P.: samgeo: A python package for segmenting geospatial data with the segment anything model (sam). Journal of Open Source Software8(89), 5663 (2023)

    2023

  10. [10]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

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

    2023

  11. [11]

    In: International Conference on Learning Representations, vol

    Ravi, N., Gabeur, V., Hu, Y.-T., Hu, R., Ryali, C., Ma, T., Khedr, H., R¨ adle, R., Rolland, C., Gustafson, L.,et al.: Sam 2: Segment anything in images and videos. In: International Conference on Learning Representations, vol. 2025, pp. 28085–28128 (2025) 22

    2025

  12. [12]

    arXiv preprint arXiv:2511.16719 (2025)

    Carion, N., Gustafson, L., Hu, Y.-T., Debnath, S., Hu, R., Suris, D., Ryali, C., Alwala, K.V., Khedr, H., Huang, A., et al.: Sam 3: Segment anything with concepts. arXiv preprint arXiv:2511.16719 (2025)

    Pith/arXiv arXiv 2025

  13. [13]

    Xu, N., Ma, X.: Llm the genius paradox: A linguistic and math expert’s struggle with simple word-based counting problems. In: Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pp. 3344–3370 (2025)

    2025

  14. [14]

    In: 2024 IEEE International Conference on Multimedia and Expo Workshops (ICMEW), pp

    Wang, M., Yuan, S., Li, Z., Zhu, L., Buys, E., Gong, M.: Language-guided zero-shot object counting. In: 2024 IEEE International Conference on Multimedia and Expo Workshops (ICMEW), pp. 1–6 (2024). IEEE

    2024

  15. [15]

    arXiv preprint arXiv:2310.11441 (2023)

    Yang, J., Zhang, H., Li, F., Zou, X., Li, C., Gao, J.: Set-of-mark prompting unleashes extraordinary visual grounding in gpt-4v. arXiv preprint arXiv:2310.11441 (2023)

    Pith/arXiv arXiv 2023

  16. [16]

    arXiv preprint arXiv:2412.00686 (2024)

    Qharabagh, M.F., Ghofrani, M., Fountoulakis, K.: Lvlm-count: Enhancing the counting ability of large vision- language models. arXiv preprint arXiv:2412.00686 (2024)

    arXiv 2024

  17. [17]

    Advances in neural information processing systems36, 19769–19782 (2023)

    Zou, X., Yang, J., Zhang, H., Li, F., Li, L., Wang, J., Wang, L., Gao, J., Lee, Y.J.: Segment everything everywhere all at once. Advances in neural information processing systems36, 19769–19782 (2023)

    2023

  18. [18]

    IEEE transactions on pattern analysis and machine intelligence 44(3), 1623–1637 (2020)

    Ranftl, R., Lasinger, K., Hafner, D., Schindler, K., Koltun, V.: Towards robust monocular depth estimation: Mix- ing datasets for zero-shot cross-dataset transfer. IEEE transactions on pattern analysis and machine intelligence 44(3), 1623–1637 (2020)

    2020

  19. [19]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Yang, L., Kang, B., Huang, Z., Xu, X., Feng, J., Zhao, H.: Depth anything: Unleashing the power of large-scale unlabeled data. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 10371–10381 (2024)

    2024

  20. [20]

    arXiv preprint arXiv:2511.10647 (2025)

    Lin, H., Chen, S., Liew, J., Chen, D.Y., Li, Z., Shi, G., Feng, J., Kang, B.: Depth anything 3: Recovering the visual space from any views. arXiv preprint arXiv:2511.10647 (2025)

    Pith/arXiv arXiv 2025

  21. [21]

    In: International Conference on Learning Representations, vol

    Bochkovskiy, A., Delaunoy, A., Germain, H., Santos, M., Zhou, Y., Richter, S., Koltun, V.: Depth pro: Sharp monocular metric depth in less than a second. In: International Conference on Learning Representations, vol. 2025, pp. 75602–75637 (2025)

    2025

  22. [22]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    G¨ ultekin, F., Koz, A., Bahmanyar, R., Azimi, S.M., S¨ uzen, M.L.: Fusing convolution and vision transformer encoders for object height estimation from monocular satellite and aerial images. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 3709–3718 (2025)

    2025

  23. [23]

    IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing19, 1883–1896 (2025)

    Li, Z., Yan, Y., Ge, B., He, Y., Zhang, Z., Zhang, Q., Gao, F.: Multistage adaptive feature enhancement mecha- nism driven single-view satellite image height estimation network for building 3-d reconstruction. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing19, 1883–1896 (2025)

    2025

  24. [24]

    https://huggingface.co/numind/NuExtract-2.0-8B

    NuMind: NuExtract 2.0 8B. https://huggingface.co/numind/NuExtract-2.0-8B. Hugging Face model repository. Accessed: 2026-07-24 (2026)

    2026

  25. [25]

    arXiv preprint arXiv:2604.08541 (2026)

    Xu, H., Hong, H., Li, H., Zhou, R., Zhang, Y., Huang, L., Xue, H., Shen, Y., Lu, W., Zhuang, Y.: Seeing but not thinking: Routing distraction in multimodal mixture-of-experts. arXiv preprint arXiv:2604.08541 (2026)

    Pith/arXiv arXiv 2026

  26. [26]

    X (formerly Twitter)

    Government of Iran: Official Statement on Country-Wide Structural Damage and Building Unit Metrics. X (formerly Twitter). Accessed: June 2026 (2026). https://x.com/Iran GOV/status/2035805820324094299

    arXiv 2026

  27. [27]

    NRC News

    Norwegian Refugee Council: Iran: One month of war leaves millions in extreme uncertainty. NRC News. Accessed: June 2026 (2026). https://www.nrc.no/news/2026/ iran-one-month-of-war-leaves-millions-in-extreme-uncertainty

    2026

  28. [28]

    The Collected Papers of Bertram Hopkinson (1915)

    Hopkinson, B.: British ordnance board minutes 13565: Explosions of high explosives. The Collected Papers of Bertram Hopkinson (1915)

    1915

  29. [29]

    Springer1(1926)

    Cranz, C.: Lehrbuch der ballistik. Springer1(1926)

    1926

  30. [30]

    In: Nordic LS-DYNA Users’ Conference, Gothenburg, Sweden (2016)

    Hilding, D.: Methods for modelling air blast on structures in ls-dyna. In: Nordic LS-DYNA Users’ Conference, Gothenburg, Sweden (2016)

    2016

  31. [31]

    JOURNAL OF BUILDING ENGINEERING124(2026) 23

    Zhou, Q., Yan, Q., Lyu, C.: Damage assessment of reinforced concrete and bonded steel strengthening rein- forced concrete beams under blast loads based on scaled distance-rotation curves. JOURNAL OF BUILDING ENGINEERING124(2026) 23

    2026

  32. [32]

    https://en.defence-ua.com/ weapon and tech/ukrainian drone makers reveal new kh s7 fpv drone and announce multirole kh s10 attack uav too-7883.html

    Defense Express: Ukrainian Drone Makers Reveal New KH-S7 FPV Drone (2023). https://en.defence-ua.com/ weapon and tech/ukrainian drone makers reveal new kh s7 fpv drone and announce multirole kh s10 attack uav too-7883.html

    2023

  33. [33]

    https://en.missilery.info/missile/grad/ shema

    Missilery.info: M-21OF Rocket (9M22U) Technical Specifications (2024). https://en.missilery.info/missile/grad/ shema

    2024

  34. [34]

    https: //missilethreat.csis.org/missile/shahed-131-and-136/

    Missile Defense Project, Center for Strategic and International Studies: Shahed-131 and -136 (2026). https: //missilethreat.csis.org/missile/shahed-131-and-136/

    2026

  35. [35]

    https://www

    Army Recognition: Shahed-136 Loitering Munition Technical Data (2024). https://www. armyrecognition.com/military-products/army/unmanned-systems/unmanned-aerial-vehicles/ shahed-136-loitering-munition-kamikaze-suicide-drone-technical-data

    2024

  36. [36]

    https://www.globalsecurity.org/military/systems/munitions/ m31.htm

    GlobalSecurity.org: M31 GMLRS Unitary (2024). https://www.globalsecurity.org/military/systems/munitions/ m31.htm

    2024

  37. [37]

    https://www.gdots.com/ missiles-and-rockets/warheads-and-payloads/gmlrs/

    General Dynamics Ordnance and Tactical Systems: GMLRS Unitary Warhead (2024). https://www.gdots.com/ missiles-and-rockets/warheads-and-payloads/gmlrs/

    2024

  38. [38]

    https://metis.fenixinsight.com/munition/rocket/fadjr-5

    Fenix Insight Ltd.: Fadjr-5 Rocket. https://metis.fenixinsight.com/munition/rocket/fadjr-5. METIS Ordnance Database. Accessed: 2026-05-29 (2025)

    2026

  39. [39]

    https://en.wikipedia.org/wiki/Fajr-5

    Wikipedia contributors: Fajr-5. https://en.wikipedia.org/wiki/Fajr-5. Accessed: 2026-05-29 (2026)

    2026

  40. [40]

    https:// missilethreat.csis.org/missile/hsiung-feng-ii/

    Missile Defense Project, Center for Strategic and International Studies: Hsiung Feng II (2024). https:// missilethreat.csis.org/missile/hsiung-feng-ii/

    2024

  41. [41]

    https: //missilethreat.csis.org/missile/kh-101-kh-102/

    Missile Defense Project, Center for Strategic and International Studies: Kh-101 / Kh-102 (2024). https: //missilethreat.csis.org/missile/kh-101-kh-102/

    2024

  42. [42]

    https://www.armyrecognition.com/military-products/army/missiles/ bombs/mk-84-bomb

    Army Recognition: Mk-84 bomb (2026). https://www.armyrecognition.com/military-products/army/missiles/ bombs/mk-84-bomb

    2026

  43. [43]

    https://www.globalmilitary.net/ bombs/fab-1500/

    GlobalMilitary.net: F AB-1500: Unguided Bomb Specs & Operators (2026). https://www.globalmilitary.net/ bombs/fab-1500/

    2026

  44. [44]

    Missile Defense Advocacy Alliance: Missile Payload & Destruction Cost Comparisons (2024). https://www.missiledefenseadvocacy.org/missile-threat-and-proliferation/future-missile-threats/ missile-payload-destruction-cost-comparisons/ Declarations • F unding:No funding was received for conducting this study. • Conflict of interest/Competing interests:The aut...

    arXiv 2024

  45. [45]

    several sites

    TARGET & SCOPE: Identify the specific proper noun or generic facility. - SCOPE CHECK: Is this a single distinct target, or does the text explicitly mention "several sites", "multiple areas", or a broad region? (Set multiple_sites_flag to "True" or "False")

  46. [46]

    Hit Ground

    GROUND IMPACT & STRUCTURAL DAMAGE: - STEP A (IMPACT STATUS): Classify into ONE of four strict statuses: * "Hit Ground": Detonated on a ground target. * "Debris": Intercepted, but falling debris impacted ground/caused fire. * "Intercepted Safe": Shot down mid-air with NO reported damage. * "Hit Ground (Assumed)": Fired towards an area; assume it hit. - STE...

  47. [47]

    a swarm",

    WEAPON CLASSIFICATION: - STEP C (WEAPON IDENTIFICATION & QUANTITIES): Identify ALL weapon types mentioned and deduce their quantities. If a specific number isn’t provided (e.g., "a swarm", "several", "a barrage"), estimate a logical integer based on the context. If singular, use 1. - Classify EACH identified weapon into one of these strict categories: * "...

  48. [48]

    Count a building if any part of its roof touches or is inside the red ring

  49. [49]

    Skip non-building objects: trees, vehicles, shadows, and debris/ruins

  50. [50]

    VERDICT DEFINITIONS: - FULLY_INSIDE: The entire roof structure is within the red boundary

    Treat attached structures as a single building unless a visible gap or parapet wall clearly separates them. VERDICT DEFINITIONS: - FULLY_INSIDE: The entire roof structure is within the red boundary. - PARTIALLY_INSIDE: The red boundary line visibly passes through the roof structure. - TOTAL_IMPACTED: The sum of the above. After your analysis, output this ...

  51. [51]

    Google Satellite (oblique) –- 3D confirmation

  52. [52]

    Buildings_Fully_Inside

    ESRI Satellite (nadir) –- primary for all boundary decisions Count every intact building roof intersecting the red-shaded zone. Skip trees, cars, shadows, ruins. Split only if a visible gap or parapet separates them. Verdict definitions: Fully Inside : entire roof under red shading (touching boundary = Inside) Partially Inside: red boundary visibly cuts t...