Multithreaded decoding and transforms of jpg strings
pip install .
- boost C++ libraries
- numpy
- PIL (for batch decoding)
The extension module _ImWorker.so provides a method:
decodeTransformListMT(jpglist=jpglist, tgt=tgt, orig_size=orig_size,
crop_size=crop_size, center=center, rgb=rgb,
flip=flip, nthreads=nthreads, calcmean=docalcmean)
jpglist: list Contains the list of coded jpg stringstgt: numpy array Must be pre-allocated to hold the decoded array data- if
calcmeanis True:tgtwill hold the mean of the images injpglistupon completiontgtmust be KxKxC, where K isorig_sizeand C is 3 when RGB is True and C is 1 otherwise.
- if calcmean is False:
tgtwill hold the decoded images in thejpglistupon completiontgtmust be N x (KxKxC), where K and C are as above and N is the number of images in jpglist.
- in both cases,
tgtmust be of type numpy.uint8
- if
orig_size: int Input dimension for decoded image. Assumes that the coded jpg images have already been pre-processed to have uniform size. If images are 256x256 for example,orig_sizeis 256.crop_size: int Desired output dimension for the decoded/cropped image. Only applies whencalcmeanis Falsecenter: bool, optional Whether to take just the center crop of every image when doing decode. Default is False.flip: bool, optional Whether to randomly flip the images horizontally. Default is False.rgb: bool, optional Whether to decode the images into RGB color space or grayscale. Default is True.nthreads: int, optional How many threads to use for decoding. Default is 1.calcmean: bool, optional Whether to calculate the mean across all images and store intgtor just do a decode and transform of images and store each individual decoded image as rows intgt. Default is False