https://github.com/dnoursi/img-instance-rec
This software package contains code for recognition of image instances within large scene images. We implement algorithms for descriptor vector construction, and scalable nearest neighbors with a BBF k-d tree.
The API can be called through a single function, detect_instances(scene, instances), within instance_detection.py. This function takes two arguments, a scene image in which to detect object instances, and a list of object instance images.
Run a demonstration with the following shell command:
$ python3 instance_detection.py
In this demonstration, an instance of the image of a coke bottle is recognized within a large scene containing furniture, humans, food, and drinks. The demonstration will display the detected image instances, as well as the full scene image.
The code implementations of algorithms for detecting interest points, generating features, and executing nearest neighors is within features.py and canny.py.
Helper code for handling image I/O with various third party libraries is primarily in util.py, visualize.py, and gen_data.py .
instance_detection.py contains the primary high-level API function detect_instances(scene, instances), as well as code for running the demonstration as described above.