A repository for creating small defects on grayscale images. The code also features slight augmentations (rotation, color change, pad and crop). The method distort_image return the altered images and the list of bouding boxes in imgaug and yolo formats.
All the functions can be found in image_defects_script.py. Notebook image_defects_imgaug.ipynb contains visualizations of the code.
Sample code:
img = cv2.imread('logo.jpg',0)
mask = ...
img_dist, bb_list, yolo_bbs = distort_image(img, mask)
Mask is an array with the same shape as the image, with 1's denoting the logo on the image and 0's denoting th e background. It can be loaded from an external source (it's just a numpy array) or can be naively generated in the following manner:
PATH_TO_IMG = 'logo.jpg'
MASK_THRESHOLD = 25
img = cv2.imread(PATH_TO_IMG, 0)
mask = np.full(img.shape, fill_value=0.0, dtype=np.float32)
for x in range(img.shape[1]):
for y in range(img.shape[0]):
if img[y, x] <= MASK_THRESHOLD:
mask[y, x] = 1
MASK_THRESHOLD is the color threshold to destinguish between the black logo and white background.
More examples and visualizations can be found in the notebook.
Sample output:
