This is an implementation of SIFT done entirely in Python with the help of NumPy.
requirement
Python 2.7.15
opencv-python==3.4.2.16 (if testing on flann would use it)
numpy==1.16.2
scipy, PIL
Run the quickly demo
python main.py
All results use brute force feature matching
SIFT_THRESHOLD = 3, 1757 features are extracted at each image
SIFT_THRESHOLD = 5, 900 features are extracted at each image
SIFT_THRESHOLD = 10, 351 features are extracted at each image
All results use SIFT_THRESHOLD=5
Brute force w/ constraints
Brute force w/o constraints
flann w/ constraints
flann force w/o constraints
