samplefilecreator is used to create the bg.txt and info.data files used by opencv_createsamples. It is also used to resize the positive images to a mean size or an user defined size, resized image are stored in a new dir called pos_resized. For creating samples it depends on mergevec. Just download it and extract it to the CascadeImgScript folder.
Currently the script assumes the dir structure you see below.
$ # The script is written for Python 3
$ python samplefilecreator/samplefilecreator.py --help
usage: samplefilecreator.py [--help] [-p [PATH]] [--jpg] [--png] [--neg]
[--pos] [-r] [-w [WIDTH]] [-h [HEIGHT]] [-m] [-v]
[--num [NUM]] [-mx [MAXXANGLE]] [-my [MAXYANGLE]]
[-mz [MAXZANGLE]]
Directory structure:
/dir
/neg
img1.jpg
img2.jpg
/pos
img1.jpg
img2.jpg
/vec
1.vec
2.vec
bg.txt
info.dat
out.vec
optional arguments:
--help show this help message and exit
-p [PATH], --path [PATH]
path to dir to process or current path of script is
used
--jpg file output is jpg (default)
--png file output is png
--neg generate bg file for negative images
--pos generate file for positive images (default)
-r, --resize resize the image to a mean value on default or to a
specific width and height
-w [WIDTH], --width [WIDTH]
width of resized images
-h [HEIGHT], --height [HEIGHT]
height of resized images
-m, --mean use mean size of old images for resized images
(default)
-v, --vec create a vec file of all pos images with using all neg
images, mergevec.py needed
--num [NUM] number_of_samples, 100 is default
-mx [MAXXANGLE], --maxxangle [MAXXANGLE]
max_x_rotation_angle, default is 1.0
-my [MAXYANGLE], --maxyangle [MAXYANGLE]
max_y_rotation_angle, default is 1.0
-mz [MAXZANGLE], --maxzangle [MAXZANGLE]
max_z_rotation_angle, default is 0.1
# Assuming you're in the same folder as the README.md file
# Generate the info.dat file in the same dir
$ python samplefilecreator/samplefilcreator.py
# Generate the info.dat file in a specific dir
$ python samplefilecreator/samplefilcreator.py -p PATH_TO_DIR
# Generate the bx.txt file in a specific dir
$ python samplefilecreator/samplefilcreator.py -p PATH_TO_DIR --neg
# On default the dir is scanned for jpg images, set png flag to scan for png
$ python samplefilecreator/samplefilcreator.py -p PATH_TO_DIR --png
# Resize the images to a specific size, they are saved in pos_resized dir
$ python samplefilecreator/samplefilcreator.py -p PATH_TO_DIR -r -w WIDTH -h HEIGHT
# Create samples from all images that were resized to a mean value
# For all images opencv_createsamples is run and then they get merged in one vec file (out.vec) (dependencies: [mergevec](https://github.com/thacoon/mergevec))
$ python samplefilecreator/samplefilcreator.py -p PATH_TO_DIR -r -m --vec
# Run the tests
$ python -m unittest test.test_samplefilecreator
# For more info take a quick look at General Usage.neg/img1.pg
neg/img2.jpg
pos/img1.jpg 1 0 0 25 25
pos/img2.jpg 1 0 0 25 25