soledad89 / scrapecars Goto Github PK

How lazy can I be making a dataset of car images

car make and models derived from: https://github.com/n8barr/automotive-model-year-data
most popular car colors : http://www.forbes.com/pictures/egdh45ijhf/1-white/#413d9991b1d4
deepLearning repos: https://github.com/fchollet/deep-learning-models

-----
general workflow:
  make a datafile of all the URLS to crawl
  use datafile to download the URLS to local files
  verify the filetypes are accurate (only want images)
  verify the file contents using ResNet50 (only want images containing a specific thing..)
  generate additional classes/tags for future training
  cut dataset into training and testing
  ...
  profit