def visualize(self,
image_name: str,
image: np.ndarray,
polygons: typing.List[typing.List[float]],
score_map: np.ndarray,
training_mask: np.ndarray,
rois: np.ndarray,
transcripts: typing.List[str] = None):
polygon_list = []
for polygon in polygons:
polygon_list.append(
ia_polys.Polygon(np.array(polygon).reshape(4, 2)))
polygons_on_image = ia_polys.PolygonsOnImage(polygons=polygon_list,
shape=image.shape)
new_image = polygons_on_image.draw_on_image(image)
colors = [(255, 0, 0), (0, 255, 0), (0, 0, 255), (0, 0, 0)]
for polygon in polygons:
for i, p in enumerate(polygon.reshape(4, 2)):
cv2.circle(new_image, tuple(p), radius=5, color=colors[i])
cv2.imwrite(image_name + '_bbox.jpg', new_image[:, :, ::-1])
rois_list = []
new_image = image.copy()
for roi, transcript in zip(rois, transcripts):
center = (roi[0], roi[1])
wh = (roi[3], roi[2])
angle = -roi[4]
box = cv2.boxPoints((center, wh, angle))
rois_list.append(ia_polys.Polygon(box))
origin = box[0]
font = cv2.FONT_HERSHEY_PLAIN
new_image = cv2.putText(new_image, transcript,
(int(origin[0]), int(origin[1] + 10)),
font, 1, (0, 255, 0), 1)
rois_on_image = ia_polys.PolygonsOnImage(polygons=rois_list,
shape=image.shape)
new_image = rois_on_image.draw_on_image(new_image)
cv2.imwrite(image_name + '_rois.jpg', new_image[:, :, ::-1])
score_map = ia_segmaps.SegmentationMapsOnImage(
score_map.astype(dtype=np.uint8), shape=image.shape)
new_image = score_map.draw_on_image(image.astype(dtype=np.uint8))
cv2.imwrite(image_name + '_score.jpg', new_image[0][:, :, ::-1])
training_mask = cv2.resize(training_mask,
dsize=(image.shape[1], image.shape[0]))
new_image = image * (1 - training_mask[:, :, None])
cv2.imwrite(image_name + '_mask.jpg', new_image[:, :, ::-1])
def test_legacy_support_for_float32_2d(self):
arr = np.array([0.4, 0.6], dtype=np.float32).reshape((1, 2))
with warnings.catch_warnings(record=True) as caught_warnings:
warnings.simplefilter("always")
segmap = segmapslib.SegmentationMapsOnImage(arr, shape=(1, 1, 3))
arr_expected = np.array([0, 1], dtype=np.int32).reshape((1, 2, 1))
assert np.array_equal(segmap.arr, arr_expected)
assert segmap.shape == (1, 1, 3)
assert len(caught_warnings) == 1
assert ("Got a float array as the segmentation map in"
in str(caught_warnings[0].message))
def basic_image_augmentation(image_name, image_dataframe):
""" Basic image augmentation """
# Basic augmentation sequence:
seq = iaa.Sequential([
iaa.Crop(px=(0, 16)),
iaa.Fliplr(0.5),
iaa.GaussianBlur(sigma=(0, 3.0)),
iaa.Rotate((-45, 45))
])
# Open image:
image_path = '../data/train_images/' + image_name
image = cv2.imread(image_path)
# Get image's mask:
image_rle = image_dataframe[(
image_dataframe['ImageId'] == image_name)]['EncodedPixels'].values[0]
image_mask = rle_to_mask(image_rle)
# Create segmentation map:
segmap = np.zeros((image.shape[0], image.shape[1]), dtype=np.int32)
segmap = np.where(image_mask == 1, 1, segmap)
segmap = iaas.SegmentationMapsOnImage(segmap, shape=image.shape)
# Apply basic augmentation sequence:
image_aug, segmap_aug = seq(image=image, segmentation_maps=segmap)
# Extract mask:
image_aug_mask = segmap_aug.get_arr()
# Get rle:
image_aug_rle = mask_to_rle(image_aug_mask)
# Return results:
return image_aug, image_aug_mask, image_aug_rle