by Pierre Sermanet (January 10th, 2011)
dscompile assembles preprocessed samples for training and testing purposes. It accepts a variety of input format, from simple directory structure to Xml files describing objects bounding boxes (e.g. PASCAL VOC format). See an example of sample extraction in this video.
Output files (datasets/images) are saved by default in the calling directory or in the directory specified via '-out' option. The outputs can be saved either as single-file datasets (default) or individual files for each sample (can be set via '-save' option).
When saving a dataset in single-file mode, the following files are created (with a dataset named 'ds'):
./dscompile <root> -jitter 1,2,3,.4,5,60,7
./dscompile <root> -jitter 0,0,0,0,30,60,7
inria dsprepare shell scripts
inria dsprepare python scripts
channels are: RGB (default), YpUV, HSV, Yp (Yp only in YpUV)
No preprocessing, i.e. no resizing or conversion.
Sleep between frame display.
Separate classes with pose if available.
This loads the dataset instead of compiling it from images found in root.
Exclude inputs for which one dimension is less than specified.
Multiply bounding boxes by a factor.
Multiply bboxes height by a factor.
Multiply bboxes width by a factor.
Force w to be h * this factor.
Include all but excluded classes, exclude can be called multiple times.
Exclude all but included classes, include can be called multiple times.
Also extract object parts, e.g. person→(head,hand,foot.
Only extract object parts, e.g. person→(head,hand,foot.
Ignore sample if “difficult” flag is on.
Ignore sample if “truncated” flag is on.
Ignore sample if “occluded” flag is on.
Ignore padded image too small for target size.
Add n samples randomly jittered from spatial neighborhood hxw, nscales within scale_range and nrotations within rotation range around original location/scale)
Add mirrored sample using vertical-axis symmetry.