pith. sign in

arxiv: 1907.05285 · v1 · pith:QRYS6LOBnew · submitted 2019-06-30 · 💻 cs.CV

Towards fully automated post-event data collection and analysis: pre-event and post-event information fusion

Ali Lenjani , Shirley J. Dyke , Ilias Bilionis , Chul Min Yeum , Kenzo Kamiya , Jongseong Choi , Xiaoyu Liu , Arindam G. Chowdhury This is my paper

Pith reviewed 2026-05-25 13:07 UTC · model grok-4.3

classification 💻 cs.CV
keywords post-event reconnaissancebuilding damage assessmentconvolutional neural networksscene classificationprobabilistic fusionstructural conditionimage analysis
0
0 comments X

The pith

Fusing pre-event and post-event images automates building categorization and post-disaster condition assessment.

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

The paper develops an automated technique to support post-event reconnaissance missions by rapidly categorizing buildings according to key physical attributes and assessing their structural condition. It processes data through separate pre-event and post-event streams that apply convolutional neural networks to classify scenes from multiple images. A probabilistic fusion step then combines those classifications to produce robust decisions about each target building. The approach is validated on images collected after hurricanes Harvey and Irma.

Core claim

The technique divides analysis into pre-event and post-event streams that each extract information about target buildings from images; CNN-based scene classification is performed on individual images and a probabilistic method fuses the results across images to decide on building attributes and post-event condition.

What carries the argument

Dual pre-event and post-event streams that run CNN scene classification on images followed by probabilistic fusion of per-image results to reach a decision on each building.

If this is right

  • The preliminary survey step of reconnaissance can be performed with less manual observation and note-taking.
  • Building attributes and structural condition can be extracted automatically from existing image collections.
  • Reconnaissance teams receive prioritized locations for detailed surveys based on the fused classifications.
  • The same pipeline can be applied to data from future events once matching between image sets is available.

Where Pith is reading between the lines

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

  • The method could be extended to include additional sensor data beyond visible images if matching to buildings remains reliable.
  • Integration with mapping tools would let teams visualize the categorized buildings directly on a geographic interface.
  • Performance on new event types would test whether the CNN classifiers require retraining or only the fusion weights need adjustment.

Load-bearing premise

Pre-event and post-event images can be reliably associated with the same target buildings so that the two streams can be fused.

What would settle it

A collection of pre-event and post-event image pairs where buildings cannot be matched across the two streams, causing the fusion step to produce incorrect attribute or condition labels.

Figures

Figures reproduced from arXiv: 1907.05285 by Ali Lenjani, Arindam G. Chowdhury, Chul Min Yeum, Ilias Bilionis, Jongseong Choi, Kenzo Kamiya, Shirley J. Dyke, Xiaoyu Liu.

Figure 1
Figure 1. Figure 1: Diagram showing the steps in the automated procedure. ical attributes to be used for the preliminary screening, as well as several pre-event views of the building from various perspectives. These two sets of complementary information are organized in a way that assists the decision-making pro￾cess of human inspectors regarding where to focus resources during a detailed survey. For clarity, we design a clas… view at source ↗
Figure 2
Figure 2. Figure 2: Hierarchy of classifiers used in pre-event and post-event data analysis streams [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Detailed steps in the post-event data analysis stream. such unstructured and complex data, as is often the case in reconnaissance datasets. Thus, a clear definition for each class is needed to establish consistent ground-truth data that are suitable for training. The definitions for those compris￾ing the post-event and pre-event streams are discussed in the following sections. 2.1.1. Classifiers used in th… view at source ↗
Figure 4
Figure 4. Figure 4: Samples of images classified as overview (OV) and non-overview (NOV). • Non-overview (hereafter, NOV): Images that are not OV are NOV. Examples of NOV include images of the interior of the building, measurements, GPS de￾vices, drawings, multiple buildings, building facades occluded by trees, cars or other buildings. Samples of images defined as OV and NOV are shown in Figs. 4a and 4b, respectively. Next, a… view at source ↗
Figure 5
Figure 5. Figure 5: Samples of images classified as major damage (MD) and non-major damage (NMD) [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Detailed steps in the pre-event data analysis stream. building images from typical street view panoramas [17, 34]. We design three independent classifiers, shown in, Fig. 2a, to label the scenes containing each view of the pre-event target building. These classifiers detect: first floor eleva￾tion, number of stories, and construction material. To suc￾cessfully train the classifiers to detect building attri… view at source ↗
Figure 7
Figure 7. Figure 7: Samples of images classified as Elevated and Non-elevated building images. (a) One Story (b) Two Stories [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Samples of images classified as (a) One-story and (b) Two-stories. Here, going from the first to the second step we assumed that the raw data 1 ,…, do not provide any additional in￾formation about the building label if image labels1 ,…, are known. This assumption is discussed again in Sec. 2.2.1. For the next steps, we use the sum rule of probability, and observe that the ’s are independent conditional on … view at source ↗
Figure 9
Figure 9. Figure 9: Samples of images classified as (a) Wood and (b) Masonry. with  = {0, 1} with = 1 corresponding to the positive detection of an attribute and = 0 to detection of the alter￾native. Finally, let  be the set of possible decisions that are available to us with regard to a given building, and one void class, here called No Decision (ND), added to skip making a decision when a confident decision is not availab… view at source ↗
Figure 10
Figure 10. Figure 10: Post-event reconnaissance dataset collected after Harvey and Irma and published on DesignSafe-CI and Fulcrum [8, 26, 27]. 0 20 40 60 80 100 Epoch 0.0 0.2 0.4 0.6 0.8 1.0 Accuracy Overview classifier Damage classifier (a) Accuracy of the classifiers for post-event stream. 0 20 40 60 80 100 Epoch 0.0 0.2 0.4 0.6 0.8 1.0 Accuracy Number-of-stories classifier Elevation classifier Material classifier (b) Accur… view at source ↗
Figure 11
Figure 11. Figure 11: Accuracy plots of classifiers. 3.1.1. Sample results First, consider the case in which all of the buildings are assumed to be captured adequately with the images avail￾able, = 1 for all ≥ 1, and pick a loss function with 1 = 2 = 0.3. This choice of the loss function making mis￾takes has a unit cost, while not deciding costs thirty percent of the mistake cost. In [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Density of the fusion predictive probabilities corresponding to each different decision, assuming all buildings are sufficiently covered (with = 1 for ≥ 1) [PITH_FULL_IMAGE:figures/full_fig_p013_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Sample of a correct MD detection [PITH_FULL_IMAGE:figures/full_fig_p014_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Sample of a ND building categorization. can be interpreted as an indication that human data collec￾tors typically have an inherent bias to take fewer images of buildings with no or minor damages, or NMD buildings. The human collectors see things that are not depicted in the im￾ages they take. For future utilization of this method, assum￾ing the collected dataset contains more images of the target building… view at source ↗
Figure 15
Figure 15. Figure 15 [PITH_FULL_IMAGE:figures/full_fig_p014_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Fusion probabilities. 1 0.1 0.3 0.5 0.7 0.9 2 0.1 0.3 0.5 0.7 0.9 Accuracy 0.70 0.75 0.80 0.85 0.90 0.95 (a) Loss function parameters on the accuracy. 1 0.1 0.3 0.5 0.7 0.9 2 0.1 0.3 0.5 0.7 0.9 ND rate 0.00 0.05 0.10 0.15 0.20 0.25 (b) Loss function parameters on the ND rate [PITH_FULL_IMAGE:figures/full_fig_p015_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Accuracy vs ND rate. ure 17a demonstrates the effect of loss function parameters on the accuracy of the post-event buildings overall damage categorization. According to, Fig. 17a, decreasing both the parameters 1 and 2 , results in higher accuracy. However according to, Fig. 17b, decreasing 1 and 2 , results in a high ND rate, rate of ND predictions over all permissible predictions. To explain it more cle… view at source ↗
read the original abstract

In post-event reconnaissance missions, engineers and researchers collect perishable information about damaged buildings in the affected geographical region to learn from the consequences of the event. A typical post-event reconnaissance mission is conducted by first doing a preliminary survey, followed by a detailed survey. The preliminary survey is typically conducted by driving slowly along a pre-determined route, observing the damage, and noting where further detailed data should be collected. This involves several manual, time-consuming steps that can be accelerated by exploiting recent advances in computer vision and artificial intelligence. The objective of this work is to develop and validate an automated technique to support post-event reconnaissance teams in the rapid collection of reliable and sufficiently comprehensive data, for planning the detailed survey. The technique incorporates several methods designed to automate the process of categorizing buildings based on their key physical attributes, and rapidly assessing their post-event structural condition. It is divided into pre-event and post-event streams, each intending to first extract all possible information about the target buildings using both pre-event and post-event images. Algorithms based on convolutional neural network (CNNs) are implemented for scene (image) classification. A probabilistic approach is developed to fuse the results obtained from analyzing several images to yield a robust decision regarding the attributes and condition of a target building. We validate the technique using post-event images captured during reconnaissance missions that took place after hurricanes Harvey and Irma. The validation data were collected by a structural wind and coastal engineering reconnaissance team, the National Science Foundation (NSF) funded Structural Extreme Events Reconnaissance (StEER) Network.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a two-stream automated technique for post-event reconnaissance that uses CNNs for scene classification on pre-event and post-event imagery, followed by a probabilistic fusion step to determine building physical attributes and structural condition. The goal is to accelerate preliminary surveys; validation is claimed on StEER post-event photos from Hurricanes Harvey and Irma.

Significance. If the per-building matching and fusion can be shown to operate reliably, the work could reduce manual effort in disaster reconnaissance by providing automated attribute and damage categorization. The approach relies on off-the-shelf CNNs and generic probabilistic fusion, so its value would lie in the end-to-end integration and demonstrated performance on real reconnaissance data.

major comments (2)
  1. [Abstract] Abstract: the claim that the method 'was validated on real images from hurricanes Harvey and Irma' is unsupported because no performance metrics, error bars, accuracy figures, or confusion matrices are reported, leaving the effectiveness of the fusion step without quantitative evidence.
  2. [Validation] Validation description: the central fusion step presupposes that pre-event and post-event images can be reliably paired to the same target buildings, yet no method, accuracy, or scale of this matching (GPS, registration, or otherwise) is described or evaluated on the StEER data, which is load-bearing for the claimed automation benefit.
minor comments (1)
  1. [Abstract] Abstract and methods: expand the description of how pre-event imagery is sourced and how images are associated with specific buildings before fusion is applied.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the method 'was validated on real images from hurricanes Harvey and Irma' is unsupported because no performance metrics, error bars, accuracy figures, or confusion matrices are reported, leaving the effectiveness of the fusion step without quantitative evidence.

    Authors: We agree that the abstract's validation claim would be strengthened by quantitative evidence. The manuscript describes application to StEER imagery from Harvey and Irma but does not report accuracy figures or confusion matrices for the CNN classification or fusion steps. We will revise the abstract for accuracy and add the requested metrics, error analysis, and matrices to the validation section. revision: yes

  2. Referee: [Validation] Validation description: the central fusion step presupposes that pre-event and post-event images can be reliably paired to the same target buildings, yet no method, accuracy, or scale of this matching (GPS, registration, or otherwise) is described or evaluated on the StEER data, which is load-bearing for the claimed automation benefit.

    Authors: The manuscript assumes that pre- and post-event images correspond to the same target buildings, as is typical for StEER reconnaissance datasets that include geo-tagged imagery of specific structures. However, we acknowledge that no explicit pairing method, accuracy evaluation, or scale is described. We will add a description of the geo-referencing approach used for pairing along with its limitations and any manual verification steps performed. revision: yes

Circularity Check

0 steps flagged

No circularity detected; derivation relies on external CNNs and generic probabilistic fusion without self-referential reductions.

full rationale

The paper presents a technique that applies standard convolutional neural networks for scene classification on pre-event and post-event images, followed by a generic probabilistic fusion step to combine results for building attributes and condition. No equations, derivations, or first-principles results are described that reduce by construction to fitted parameters or inputs defined from the same data. The approach cites external CNN models and does not invoke self-citations as load-bearing uniqueness theorems or ansatzes. Validation is performed on external StEER post-event imagery from hurricanes Harvey and Irma. The central claims rest on independent components (pre-trained CNNs and standard fusion) rather than tautological steps, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the applicability of off-the-shelf CNNs to disaster imagery and on the ability to associate pre- and post-event images with the same buildings; no new entities are postulated and the only free parameters are those internal to the trained CNNs.

free parameters (1)
  • CNN weights
    Learned during training on image classification tasks; specific training corpus and hyperparameters not stated in abstract.
axioms (2)
  • domain assumption Convolutional neural networks trained on general scene data produce usable classifications of building attributes in post-disaster imagery
    Invoked when the abstract states that CNNs are implemented for scene classification without domain-specific retraining details.
  • domain assumption Multiple independent image classifications of the same building can be combined via a probabilistic model to yield a robust label
    Stated directly in the description of the fusion step.

pith-pipeline@v0.9.0 · 5844 in / 1331 out tokens · 30847 ms · 2026-05-25T13:07:47.739199+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [1]

    Googlestreetview: Captur- ing the world at street level

    Anguelov, D., Dulong, C., Filip, D., Frueh, C., Lafon, S., Lyon, R., Ogale,A.,Vincent,L.,Weaver,J.,2010. Googlestreetview: Captur- ing the world at street level. Computer 43, 32–38

    work page 2010

  2. [2]

    Computer-aided approach for rapid post-event visual evaluation of a building façade

    Choi, J., Yeum, C., Dyke, S., Jahanshahi, M., 2018. Computer-aided approach for rapid post-event visual evaluation of a building façade. Sensors 18, 3017

    work page 2018

  3. [3]

    Xception: Deeplearningwithdepthwiseseparable convolutions, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Chollet, F., 2017. Xception: Deeplearningwithdepthwiseseparable convolutions, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 1251–1258

    work page 2017

  4. [4]

    Keras.https://keras.io

    Chollet, F., et al., 2015. Keras.https://keras.io

    work page 2015

  5. [5]

    Univ of California Press

    Comerio,M.C.,1998.Disasterhitshome: Newpolicyforurbanhous- ing recovery. Univ of California Press

    work page 1998

  6. [6]

    CONVERGE

    CONVERGE Team, 2019 (accessed: 04.06.2019). CONVERGE. URL: https://converge.colorado.edu

    work page 2019

  7. [7]

    NHERI Five-year Science Plan

    DesignSafe-CI, 2017 (accessed: 22.02.2019). NHERI Five-year Science Plan. URL:https://www.designsafe-ci.org/facilities/nco/ science-plan

    work page 2017

  8. [8]

    DesignSafe-CI: A Comprehensive Cyberinfrastructure Environment for Research in NaturalHazardsEngineering.URL: https://www.designsafe-ci.org/

    DesignSafe-CI, 2018 (accessed: 22.02.2019)a. DesignSafe-CI: A Comprehensive Cyberinfrastructure Environment for Research in NaturalHazardsEngineering.URL: https://www.designsafe-ci.org/

    work page 2018

  9. [9]

    Rapid experimental facility

    DesignSafe-CI, 2018 (accessed: 22.02.2019)b. Rapid experimental facility. URL:https://rapid.designsafe-ci.org

    work page 2018

  10. [10]

    The pascal visual object classes (voc) challenge

    Everingham, M., Van Gool, L., Williams, C.K., Winn, J., Zisserman, A., 2010. The pascal visual object classes (voc) challenge. Interna- tional journal of computer vision 88, 303–338

    work page 2010

  11. [11]

    Deep transfer learning for image- based structural damage recognition

    Gao, Y., Mosalam, K.M., 2018. Deep transfer learning for image- based structural damage recognition. Computer-Aided Civil and In- frastructure Engineering 33, 748–768

    work page 2018

  12. [12]

    Geotechnical ExtremeEventsReconnaissance

    GEER Association, Inc, 2018 (accessed: 22.02.2019). Geotechnical ExtremeEventsReconnaissance. URL: http://www.geerassociation. org

    work page 2018

  13. [13]

    Deep learning

    Goodfellow, I., Bengio, Y., Courville, A., 2016. Deep learning. MIT press

    work page 2016

  14. [14]

    Hurricane florence: Field assessment team 1 (fat-1) early ac- cess reconnaissance report (earr)

    Kijewski-Correa, T., Prevatt, D., Musetich, M., Roueche, D., Mos- alam, K., Hu, F., Salman, A., Peng, H., Gonzalez, C., Robertson, I., 2018a. Hurricane florence: Field assessment team 1 (fat-1) early ac- cess reconnaissance report (earr)

    work page

  15. [15]

    Steer-hurricanemichael: Preliminaryvirtualassessmentteam(p-vat) report

    Kijewski-Correa, T., Prevatt, D., Robertson, I., Smith, D., Mosalam, K., Roueche, D., Lichty, B., Gutierrez Soto, M., Aly, A.M., Lenjani, A., Salman, A., Alipour, A., Davis, B., Sutley, E., Miner, N., 2018b. Steer-hurricanemichael: Preliminaryvirtualassessmentteam(p-vat) report

    work page

  16. [16]

    Hurricane harvey (texas) supplement–collaborative re- search: Geotechnical extreme events reconnaissance (geer) associa- tion: Turning disaster into knowledge

    Kijewski-Correa, T., Taflanidis, A., Kennedy, A., Womble, A., Liang, D., Starek, M., Peterman, K., Sun, C., Leite, F., Wood, R., et al., 2018c. Hurricane harvey (texas) supplement–collaborative re- search: Geotechnical extreme events reconnaissance (geer) associa- tion: Turning disaster into knowledge

    work page

  17. [17]

    Auto- matedbuildingimageextractionfrom360-degreepanoramasforpost- disaster evaluation

    Lenjani, A., Yeum, C.M., Dyke, S., Bilionis, I., 2019. Auto- matedbuildingimageextractionfrom360-degreepanoramasforpost- disaster evaluation. arXiv preprint arXiv:1905.01524

    work page arXiv 2019

  18. [18]

    NaturalHazards Engineering Research Infrastructure: Post-Disaster, Rapid Response Research(RAPID)Facility

    NSFAward1611820,2016(accessed: 22.09.2018). NaturalHazards Engineering Research Infrastructure: Post-Disaster, Rapid Response Research(RAPID)Facility. URL: https://www.nsf.gov/awardsearch/ showAward?AWD_ID=1611820&HistoricalAwards=false

    work page 2016

  19. [19]

    CONVERGE: Coordinated Social Science, Engineering, and Interdisciplinary Ex- treme Events Reconnaissance Research

    NSF Award 1841338, 2019 (accessed: 22.09.2018). CONVERGE: Coordinated Social Science, Engineering, and Interdisciplinary Ex- treme Events Reconnaissance Research. URL: https://nsf.gov/ awardsearch/showAward?AWD_ID=1841338&HistoricalAwards=false

    work page 2019

  20. [20]

    EAGER: Operationalization of the Structural Extreme Events Reconnais- sance (StEER) Network

    NSF Award 1841667, 2016 (accessed: 22.09.2018). EAGER: Operationalization of the Structural Extreme Events Reconnais- sance (StEER) Network. URL: https://www.nsf.gov/awardsearch/ showAward?AWD_ID=1841667&HistoricalAwards=false

    work page 2016

  21. [21]

    Recent AwardsMadebyNSFThroughNHERIProgram

    NSF Awards 1612843, 1519679, 1840870, 1520765, 1611820, 1520817, 1520581, 1520843, 1520853, 1612144, 1520808, 1520904, and 1841338, 2019 (accessed: 22.09.2018). Recent AwardsMadebyNSFThroughNHERIProgram. URL: https://www. nsf.gov/awardsearch/advancedSearchResult?ProgEleCode=013Y,7470& BooleanElement=Any&BooleanRef=Any&ActiveAwards=true&#results

    work page 2019

  22. [22]

    Natural Hazards Engineering Re- search Infrastructure (2015 - 2019) (NHERI)

    NSF PROGRAM SOLICITATION 14-605, 2015 (ac- cessed: 22.09.2018). Natural Hazards Engineering Re- search Infrastructure (2015 - 2019) (NHERI). URL: https://www.nsf.gov/pubs/2014/nsf14605/nsf14605.htm

    work page 2015

  23. [23]

    Natural Hazards Engineering Research Infrastructure (2019) (NHERI)

    NSF PROGRAM SOLICITATION 15-598, 2019 (accessed: 22.09.2018). Natural Hazards Engineering Research Infrastructure (2019) (NHERI). URL: https://www.nsf.gov/funding/pgm_summ.jsp? pims_id=503259

    work page 2019

  24. [24]

    Overview of damage observed in regional construction during the passage of hurricane irma over the state of florida

    Pinelli, J.P., Roueche, D., Kijewski-Correa, T., Plaz, F., Prevatt, D., Zisis, I., Elawady, A., Haan, F., Pei, S., Gurley, K., et al., 2018. Overview of damage observed in regional construction during the passage of hurricane irma over the state of florida. Proc., ASCE Forensic 18

    work page 2018

  25. [25]

    Steer-hurricane michael: Field assessment team 1 (fat-1) early access reconnaissance report (earr)

    Roueche, D., Cleary, J., Gurley, K., Marshall, J., Pinelli, J.P., Pre- vatt,D.,Smith,D.,Alipour,A.,Angeles,K.,Davis,B.,Gonzalez,C., Lenjani, A., Mulchandani, H., Musetich, M., Salman, A., Kijewski- Correa, T., Robertson, I., Mosalam, K., 2018. Steer-hurricane michael: Field assessment team 1 (fat-1) early access reconnaissance report (earr)

    work page 2018

  26. [26]

    Fulcrum|Mobile FormBuilder&DataCollectionApp

    SpatialNetworks,Inc,2018(accessed: 22.09.2018). Fulcrum|Mobile FormBuilder&DataCollectionApp. URL: https://www.fulcrumapp. com

    work page 2018

  27. [27]

    NSF Structural Ex- treme Events Reconnaissance (StEER) Network

    StEER Network, 2018 (accessed: 22.09.2018). NSF Structural Ex- treme Events Reconnaissance (StEER) Network. URL:https://web. fulcrumapp.com/communities/nsf-rapid

    work page 2018

  28. [28]

    Structural Extreme Events Reconnaissance Network

    StEER Network, 2018 (accessed: 25.04.2019). Structural Extreme Events Reconnaissance Network. URL:https://www.steer.network

    work page 2018

  29. [29]

    Inception- v4, inception-resnet and the impact of residual connections on learn- ing, in: Thirty-First AAAI Conference on Artificial Intelligence

    Szegedy, C., Ioffe, S., Vanhoucke, V., Alemi, A.A., 2017. Inception- v4, inception-resnet and the impact of residual connections on learn- ing, in: Thirty-First AAAI Conference on Artificial Intelligence

    work page 2017

  30. [30]

    Re- thinking the inception architecture for computer vision, in: Proceed- ings of the IEEE conference on computer vision and pattern recogni- tion, pp

    Szegedy,C.,Vanhoucke,V.,Ioffe,S.,Shlens,J.,Wojna,Z.,2016. Re- thinking the inception architecture for computer vision, in: Proceed- ings of the IEEE conference on computer vision and pattern recogni- tion, pp. 2818–2826

    work page 2016

  31. [31]

    Virtual Earthquake Reconnaissance Team (VERT)

    The Earthquake Engineering Research Institute (EERI), 2019 (ac- cessed: 04.06.2019). Virtual Earthquake Reconnaissance Team (VERT). URL:http://www.learningfromearthquakes.org/activities/ vert

    work page 2019

  32. [32]

    Postevent reconnaissance image documentation us- ing automated classification

    Yeum,C.M.,Dyke,S.J.,Benes,B.,Hacker,T.,Ramirez,J.,Lund,A., Pujol, S., 2018a. Postevent reconnaissance image documentation us- ing automated classification. Journal of Performance of Constructed Facilities 33, 04018103

    work page

  33. [33]

    Engineering Structures 155, 16–24

    Yeum,C.M.,Dyke,S.J.,Ramirez,J.,2018b.Visualdataclassification in post-event building reconnaissance. Engineering Structures 155, 16–24

    work page

  34. [34]

    Automated detection of pre-disaster building images from google street view, in: Proceedings of The 7th World Conference on Structural Control and Monitoring, (7WCSCM 2018)

    Yeum, C.M., Lenjani, A., Dyke, S.J., Bilionis, I., 2018c. Automated detection of pre-disaster building images from google street view, in: Proceedings of The 7th World Conference on Structural Control and Monitoring, (7WCSCM 2018). Lenjani et al.:Preprint Page 17 of 17

    work page 2018