Releases: multi-template-matching/MultiTemplateMatching-Python
Replace with opencv NMS
This version replaces the original hard-coded NMS with the opencv cv2.dnn.NMSBoxes
method.
This simplifies the code and improve portability to other programing languages.
Performance-wise it is similar to potentially slightly faster.
The function signature is the same except the default scoreThreshold is now 0.5
Other changes:
- the method in
MTM.NMS
PointInRect
andcomputeIoU
have been removed (not used anymore for NMS) - tutorials updated with timeit benchmarks
Improve versioning
- don't mention any opencv in setup.py (more flexible installation)
- Define MTM version in a single place
Add computeScoreMap method
Add new convenient method computeScoreMap
No other change, no backward compatibility issue
Autocast 16-bit images to 32-bit for matchTemplate
OpenCV match template only support 8-bit or 16-bit images
In Multi-template matching, if both the images and template are 8-bit then the function matchTemplate is called as such.
In all other cases, the template and image are casted to 32-bit to call matchTemplate.
Use pandas for hits
1.5.1 fix formating
Add drawBoxesOnRGB
Add methods:
- drawBoxesOnRGB
- drawBoxesOnGray
Removed:
- drawBoxes
Ie clearer what the function produce
drawBoxesOnGray can be used to make mask for instance
Minor improvements
- format docstring in numpy style
- Improve doc in the readme
- drawBoxes now output a RGB image (initial grayscale image are duplicated as RGB)
Initial stable release
Stable release at time of publication