bounded_box_cropper.py

'''
After the bounding boxes have been drawn using `draw_bounding_boxes.py`,
every bounded box must be cropped and copied into its own image in order
to train and test the net on them. No resizing to 256x256 is necessary
because I let caffe do that.
TODO make sure caffe warps rather than crops.

The source images should be in `data/imagenet/<wnid>/images/all`,
and the cropped images will be placed in
`data/imagenet/<wnid>/images/cropped`.
'''

import gflags
from gflags import FLAGS
from flags import set_gflags

# This default wnid is for eggs
gflags.DEFINE_string('bounding_boxes_csv', None,
                     'The log that contains the bounding box coordinates')
gflags.MarkFlagAsRequired('bounding_boxes_csv')
gflags.DEFINE_string('dst', None, 'Where to store the cropped images')
gflags.MarkFlagAsRequired('dst')

from cropping_utils import get_crop_box

import csv
from PIL import Image
from os.path import dirname, abspath, join, splitext, basename
from os import system
from random import randint

ROOT = dirname(abspath(__file__))

if __name__ == '__main__':
  set_gflags()
  system('mkdir -p ' + FLAGS.dst)
  count = 0
  with open(FLAGS.bounding_boxes_csv) as csvfile:
    reader = csv.reader(csvfile)
    for row in reader:
      for i in range(1, len(row), 4):
        count += 1
        if count % 100 == 0:
          print count
        filename = row[0]
        image = Image.open(filename)
        width, height = image.size
        target = join(FLAGS.dst, splitext(basename(filename))[0])
        image.crop(get_crop_box(row, i, width, height)
           ).save(target + '_' + str(int(i/4)) + '.jpg')