GIS - Urban Perception

Safety in New York

  1. data collection

Crowd Sourcing from Place Pulse:
http://pulse.media.mit.edu/data/

  1. Data Cleaning

Filter out the scores in NY
from the raw json file

select a ROI in Manhattan

  1. Download Images

1705 / 4136 (41.22%) scores in NY

using urllib.request.urlretrieve() to download images

  1. Extrac Features
  1. SIFT
  1. Dense SIFT
  1. GSIT

python setup.py build_ext -I D:\Downloads\fftw-3.3.5-dll64 -L D:\Downloads\fftw-3.3.5-dll64


Using Anaconda Cmd


https://github.com/tuttieee/lear-gist-python

  1. compiling fftw using Lib.exe and copy to VS/Community
  1. Download pthread.h
  1. define some bugs

    define HAVE_STRUCT_TIMESPEC

    define M_PI 3.14159265358979323846

  1. remove POSIX thread

Crowd Scores

  1. Generates the Win/Lose labels using the Crowd scores
  1. Apply to the neural network to Predict the crowd Scores

a. Win or Loss (VGG, AlexNet...)

b. Convert to Scores (Ranking Network)

  1. Apply to other Cities
  1. Texton Histogram
  1. Color Histogram