bm_experiments.bm_comp_perform module¶
Simple benchmarks measuring basic computer performances
We run image registration in single thread and then in all available thread in parallel and measure the execution time.
The tested image registration scenario is as following
- load both images
- perform som simple denoising
- extract ORB features
- estimate affine transform via RANSAC
- warp and export image
Example run:
pip install --user tqdm numpy scikit-image https://github.com/Borda/BIRL/archive/master.zip
python bm_comp_perform.py -o ../output -n 3
Copyright (C) 2018 Jiri Borovec <jiri.borovec@fel.cvut.cz>
-
bm_experiments.bm_comp_perform.
_clean_images
(image_paths)[source]¶ remove temporary images
Parameters: image_paths (str) – path to images
-
bm_experiments.bm_comp_perform.
_prepare_images
(path_out, im_size=(2000, 2000))[source]¶ generate and prepare synth. images for registration
Parameters: Return tuple(str,str): paths to target and source image
-
bm_experiments.bm_comp_perform.
arg_parse_params
()[source]¶ parse the input parameters :return dict: parameters
-
bm_experiments.bm_comp_perform.
main
(path_out='', nb_runs=5)[source]¶ the main entry point
Parameters:
-
bm_experiments.bm_comp_perform.
measure_registration_parallel
(path_out, nb_iter=3, nb_workers=2)[source]¶ measure mean execration time for image registration running in N thread
Parameters: Return dict: dictionary of float values results
-
bm_experiments.bm_comp_perform.
measure_registration_single
(path_out, nb_iter=5)[source]¶ measure mean execration time for image registration running in 1 thread
Parameters: Return dict: dictionary of float values results