Combining Satellite Imagery and Open Data to Map Road Safety
Authors: Alameen Najjar, ShunÕichi Kaneko, Yoshikazu Miyanaga
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To empirically validate the proposed approach, we trained a deep model on satellite images obtained from over 647 thousand traffic-accident reports collected over a period of four years by the New York city Police Department. The best model predicted road safety from raw satellite imagery with an accuracy of 78%. We also used the New York city model to predict for the city of Denver a city-scale map indicating road safety in three levels. Compared to a map made from three years worth of data collected by the Denver city Police Department, the map predicted from raw satellite imagery has an accuracy of 73%. |
| Researcher Affiliation | Academia | Alameen Najjar, Shun ichi Kaneko, Yoshikazu Miyanaga Graduate School of Information Science and Technology, Hokkaido University, Japan najjar@hce.ist.hokudai.ac.jp, {kaneko, miya}@ist.hokudai.ac.jp |
| Pseudocode | No | The paper describes the methods used in prose but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured steps in a code-like format. |
| Open Source Code | No | The paper states it is 'Making publicly available a deep model' (Contribution 3), but it does not explicitly provide a link or statement that the source code for the methodology described in the paper is openly available. |
| Open Datasets | Yes | We used data collected in two US cities (New York and Denver), and it is summarized as follows: 647,868 traffic-accident reports collected by the NYPD over the period between March 2012 and March 20161. 1https://data.cityofnewyork.us/. 110,870 traffic-accident reports collected by the Denver city police department over the period between July 2013 and July 2016. [...] Our models were pre-trained on a generic large-scale image dataset first. Three pre-training datasets were considered: (1) Image Net (Deng et al. 2009), (2) Places205 (Zhou et al. 2014), and (3) both Image Net and Places205 datasets combined. |
| Dataset Splits | No | The paper states 'All models in this paper were trained, verified and tested on satellite images...' and 'To evaluate the learned models, we calculated the average prediction accuracy cross-validated on three random 5%/95% data splits.' However, it does not explicitly provide distinct split percentages or counts for a separate validation set, only implying its existence through 'verified'. |
| Hardware Specification | Yes | Finally, training was conducted using Caffe deep learning framework (Jia et al. 2014) running on a single Nvidia Ge Force TITAN X GPU. |
| Software Dependencies | No | The paper mentions using 'Caffe deep learning framework (Jia et al. 2014)' but does not provide specific version numbers for Caffe or any other key software libraries or dependencies. |
| Experiment Setup | Yes | Individual images have a spatial resolution of 256 256 pixels each and crawled at three different zoom levels (18, 19, and 20). Architecture: All Conv Nets used in this experiments follow the Alex Net architecture (Krizhevsky, Sutskever, and Hinton 2012). ... Reported results are obtained after 60,000 training iterations. |