annt¶
Description¶
Various tools have been developed so far for object detection tasks. However, there are no standard in annotation tools and formats and developers still write their own json or xml parser of annotation files. annt is an annotation tool that operates in the form of cloud services such as dropbox. annt provides not only simple and comfortable annotation exprience, but also powerful library for loading annotated images.
This is a documentation for python library which read images annotated with annt. you can load annotated images in a simple way and focus on the essential AI development. Also, this library has a basic build-in preprocessing functions. So you can save time to write extra code.
Examples¶
Example 1. Load annotated images¶
import annt
# annotations is list of annotation data
annotations = annt.load('./Dropbox/app/project_name')
# Display ths information of each annotation file.
for a in annotations:
image = a.image # opencv2 image array
boxes = a.boxes # list of bounding boxes
height, width, colors = image.shape # you can
for box in boxes:
# Tag information (str)
print(f'~ tag name : box.tag ~')
# You can get coordination information of the box by two methods,
# Left Top Style and Edge Style.
# Coordination information based on left top of the box. (Left-Top Style)
print(f'x : {box.x}')
print(f'y : {box.y}')
print(f'w : {box.w}')
print(f'h : {box.h}')
# Coordination information based on the distance from each edge of the image. (Edge Style)
print(f'left : {box.left}')
print(f'right : {box.right}')
print(f'top : {box.top}')
print(f'bottom : {box.bottom}')
# If you change these coordination properties, all of them will recomputed.
box.w = 300 # This operation will also change box.right property.
Example 2. Data augumentation¶
import annt
import random
# annotations is list of annotation data
annotations = annt.load('./Dropbox/App/annt/test')
sample_n = 10 # Number of samples from one image
# Display ths information of each annotation file.
augumented = []
for raw_a in annotations:
for i in range(sample_n):
# Rotate image
rot_deg = random.choice([0, 90, 180, 270, 360])
a = raw_a.rotate(rot_deg)
# Tilt image
tilt_deg = random.randint(-8, 8)
a = a.rotate(tilt_deg)
# Flip image
flip_x = random.randint(0, 1)
flip_y = random.randint(0, 1)
a = a.flip(flip_x, flip_y)
augumented.append(a)
# Show first augumented image.
augumented[0].show()
Documents¶
Core module of annt-python
This module provides functions related Boudning Box and Annotation information, which are important for handling annotation information.
-
class
Annotation
(filename, image, boxes=[])¶ Image and annotation information holder.
-
filename
: str¶ filename.
-
image
: np.ndarray¶ Image array in opencv2 format.
-
boxes
: list¶ List of box.
-
flip
(flip_x=True, flip_y=False)¶ Flip image. Thie method flip image by the axis given by argument. This method is non-destructive.
- Parameters
flip_x (bool, optional) – Whether flip with x axis. Default True.
flip_y (bool, optional) – Whether flip with y axis. Default True.
- Returns
Rotated annotate object.
- Return type
Annotate
-
resize
(width, height)¶ Resize image to the spcified size. This method is non-destructive.
- Parameters
width (int) – width
height (int) – height
- Returns
Resized annotate object.
- Return type
Annotate
-
rotate
(angle)¶ Rotate image at the specified angle. Create copy of itself and rotate. This method is non-destructive.
- Parameters
angle (int) – Rotate angle (degree).
- Returns
Rotated annotate object.
- Return type
Annotate
-
show
(max_width=500, max_height=500)¶ Display image with annotation information.
Notes
Press any key to close image window.
-
-
class
Box
(tag, iwidth, iheight, x, y, w, h)¶ Bounding box representation.
-
tag
: str¶ tag of the box.
-
x
: float¶ Upper-Left x coordination of the bounding box.
-
y
: float¶ Upper-Left y coordination of the bounding box.
-
w
: float¶ Width of the bounding box.
-
h
: float¶ Height of the bounding box.
-
top
: float¶ Distance from top.
-
bottom
: float¶ Distance from bottom.
-
left
: float¶ Distance from left.
-
right
: float¶ Distance from right.
-
-
load
(dir_path)¶ load annotation files.
- Parameters
dir_path (str) – Annotation directory path.
- Yields
Annotation – Loaded Annotation object.