Data set with new classifications for broken wires and cropping

[NotJoeMartinez]

Many of the old labels were mislabeled with broken wires, label and augment the new data set.

• [ ] Make sure the images are cropped to 480x480

  "convert {0} -crop 480x480+80+0 {1}".format(input_img, crop_name)

• [ ] After augmenting remove the images that have stretching on the outer corners

[NotJoeMartinez]

• Augmented images using the following augmentation settings

  
  def anticlockwise_rotation(image, fill_mode):
  angle = random.randint(0,180)
  return rotate(image, angle, resize=False,  cval=0, mode=fill_mode)

def clockwise_rotation(image, fill_mode): angle = random.randint(0,180) return rotate(image, -angle, resize=False, cval=0, mode=fill_mode)

def h_flip(image, fill_mode): return np.fliplr(image)

def v_flip(image, fill_mode): return np.flipud(image)

- Manually removed the stretched images
- Cropped images to 480x480
- Removed duplicates from testing data set using `fdupes`
    - I cut down the other datasets to the minimum number after running `fdupes` which was 89 images per folder
    - Kept duplicates in training folder