https://github.com/dakshvar22/image-transformer
Python library to apply transformations to images and regions of interests with images specifically using opencv
https://github.com/dakshvar22/image-transformer
computer-vision opencv python
Last synced: about 1 year ago
JSON representation
Python library to apply transformations to images and regions of interests with images specifically using opencv
- Host: GitHub
- URL: https://github.com/dakshvar22/image-transformer
- Owner: dakshvar22
- License: mit
- Created: 2018-01-07T11:17:40.000Z (over 8 years ago)
- Default Branch: master
- Last Pushed: 2018-03-21T05:40:25.000Z (over 8 years ago)
- Last Synced: 2025-03-25T09:19:01.125Z (about 1 year ago)
- Topics: computer-vision, opencv, python
- Language: Python
- Homepage:
- Size: 86.9 KB
- Stars: 5
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Image Transformer
This python library helps you apply a diverse combination of affine transformations on an image. Moreover, it gives you the flexibility to apply the exact same transformation on a set of synthetic points over the image(For eg - bounding boxes for image localization task). Such transformations are extremely helpful in AI related tasks specially where the training corpus is small and needs to be augmented with these transformed images.
Available Transforms -
1. Horizontal Flip
2. Vertical Flip
3. Translate in x
4. Translate in y
5. Rotation
6. Shear Transform
## Requirements and installation
Required Packages -
1. OpenCV - Currently tested with 2.4.11
To install, navigate to the root folder and run -
`
pip install -e .
`
## Usage
```python
# Import
from imgTransformer import Transformer
import cv2
# Instantiation
data_transformer = Transformer(transform_boxes = True)
'''
parameter transform_boxes lets you specify if you want to suppy the corresponding bounding box annotations as well and get the transformed boxes too. Defaults to False
'''
# Affine transformation parameters
tx_range = (-20,20) # Translate in x direction by n pixels where n is picked randomly from the range - (-20,20)
ty_range = (-20,20) # Translate in y direction by n pixels where n is picked randomly from the range - (-20,20)
rot_range = (-30,30) # Rotate by an angle theta where theta is picked randomly from the range - (-30,30)
shear_range = 5 # Apply a shear transform with shear coefficient being 5
# Supply these parameters to the data_transformer object
data_transformer.set_params(tx_range,ty_range,rot_range,shear_range)
# Load an image
img = cv2.imread('0.JPEG') # numpy matrix with dimensions - (height,width,channels)
'''
Specify bounding boxes(dummy box taken here)
A list of list where each bounding boxes is a list of size 4 with elements(in order) -
1. X coordinate of top left vertex
2. Y coordinate of top left vertex
3. X coordinate of bottom right vertex
4. Y coordinate of bottom right vertex
'''
boxes = [[132, 85, 251, 275]] # dummy box
# Apply the transform
img_trans, boxes_trans = data_transformer.applyTransforms(img,boxes,num_transforms = 2) # num_transforms specifies how many consecutive affine transformations should be applied.
# draw the bounding boxes for visualization
cv2.rectangle(img,(boxes[0][0],boxes[0][1]),(boxes[0][2],boxes[0][3]),color = (255,0,0))
cv2.rectangle(img_trans,(boxes_trans[0][0],boxes_trans[0][1]),(boxes_trans[0][2],boxes_trans[0][3]),color = (255,0,0))
# Store the files -
cv2.imwrite('original.JPEG',img)
cv2.imwrite('transformed.JPEG',img_trans)
```
## Result
Original
![alt text][Original]
Transformed
![alt text][Transformed]
[Original]: original.JPEG "Original"
[Transformed]: transformed.JPEG "Transformed"
The above mentioned transformations are applied completely randomly so a different run of test.py can result in a different transformed image from what is shown above.