def showdata(args):
cvd.checkresize(args.resize_imageW, args.resize_imageH,
args.resize_saliencyW, args.resize_saliencyH)
cvd.checknormalizerange(args.normalize_image,
args.normalize_saliency)
filelist = cvd.getfilelist(args.imagefolder, args.saliencyfolder,
args.listfile, args.shuffle)
# get the shape for h5py dataset
image, saliency = cvd.convertimageandmat(filelist[14], args.gray,
args.check_size,
args.resize_imageW,
args.resize_imageH,
args.resize_saliencyW,
args.resize_saliencyH,
args.normalize_image,
args.normalize_saliency)
with h5py.File('datasetmean.hdf5') as f:
mset = f.get('mean')
imagemean = mset[0]
mean_c = imagemean.mean(1).mean(1)
for i in range(imagemean.shape[0]):
imagemean[i].fill(mean_c[i])
image = image.astype(np.float32)
image -= imagemean
image = cvd.chw2hwc(image, args.gray)
image = cvd.normalizeimage(image, args.normalize_image)
saliency = cvd.l2hw(saliency, args.resize_saliencyW, args.resize_saliencyH)
plt.figure()
skimage.io.imshow(image)
plt.figure()
skimage.io.imshow(saliency)
skimage.io.show()
def createdata(args):
logger_root = logging.getLogger()
cvd.checkresize(args.resize_imageW, args.resize_imageH,
args.resize_saliencyW, args.resize_saliencyH)
cvd.checknormalizerange(args.normalize_image,
args.normalize_saliency)
filelist = cvd.getfilelist(args.imagefolder, args.saliencyfolder,
args.listfile, args.shuffle)
# get the shape for h5py dataset
image, saliency = cvd.convertimageandmat(filelist[0], args.gray,
args.check_size,
args.resize_imageW,
args.resize_imageH,
args.resize_saliencyW,
args.resize_saliencyH,
args.normalize_image,
args.normalize_saliency)
N = len(filelist)
datashape = tuple([N]) + image.shape
labelshape = tuple([N]) + saliency.shape
with h5py.File('datasetmean.hdf5') as f:
mset = f.get('mean')
imagemean = mset[0]
mean_c = imagemean.mean(1).mean(1)
for i in range(imagemean.shape[0]):
imagemean[i].fill(mean_c[i])
# write to h5py dataset
with h5py.File(args.outfile) as f:
dset = f.create_dataset(name = 'data', shape = datashape,
dtype = np.dtype(np.float32),
compression = "gzip", compression_opts = 4)
lset = f.create_dataset(name = 'label', shape = labelshape,
dtype = np.dtype(np.float32),
compression = "gzip", compression_opts = 4)
for i in range(N):
image, saliency = cvd.convertimageandmat(filelist[i],
args.gray,
args.check_size,
args.resize_imageW,
args.resize_imageH,
args.resize_saliencyW,
args.resize_saliencyH,
args.normalize_image,
args.normalize_saliency)
image = image.astype(np.float32)
image -= imagemean
image = cvd.chw2hwc(image, args.gray)
image = cvd.normalizeimage(image, args.normalize_image)
image = cvd.hwc2chw(image, args.gray)
dset[i] = image
lset[i] = saliency
if((i + 1) % 100 == 0):
logger_root.info('Processed %d files.' %(i + 1))
if((i + 1) % 100 != 0):
logger_root.info('Processed %d files.' %(i + 1))
def computerimagemean(args):
logger_root = logging.getLogger()
filelist = cvd.getfilelist(args.imagefolder, args.saliencyfolder,
args.listfile, args.shuffle)
imagelist = filelist[:,0]
image = cvd.convertimage(imagelist[0], args.gray, args.resize_imageW,
args.resize_imageH, args.normalize_image)
N = len(imagelist)
imagemean = np.zeros(image.shape, dtype = np.float64)
for i in range(N):
image = cvd.convertimage(imagelist[i], args.gray,
args.resize_imageW,
args.resize_imageH,
args.normalize_image)
imagemean += image
if((i + 1) % 100 == 0):
logger_root.info('Processed %d files.' %(i + 1))
if((i + 1) % 100 != 0):
logger_root.info('Processed %d files.' %(i + 1))
imagemean /= (N + 1)
# range [0.0 255.0] means image is uint8
if(args.normalize_image[0] == 0 and args.normalize_image[1] == 255):
return imagemean.astype(np.uint8)
return imagemean.astype(np.float32)
sys.exit(1)
if args.resize_imageH <= 0:
logger_root = logging.getLogger()
logger_root.error("resize_imageH should be greater than 0")
sys.exit(1)
if args.resize_saliencyW <= 0:
logger_root = logging.getLogger()
logger_root.error("resize_saliencyW should be greater than 0")
sys.exit(1)
if args.resize_saliencyH <= 0:
logger_root = logging.getLogger()
logger_root.error("resize_saliencyH should be greater than 0")
sys.exit(1)
filelist = convertdata.getfilelist(args.imagefolder, args.saliencyfolder, args.listfile, args.shuffle)
imageHWC, saliency = convertdata.readimageandmat(
filelist[8],
args.gray,
args.check_size,
args.resize_imageW,
args.resize_imageH,
args.resize_saliencyW,
args.resize_saliencyH,
)
imageCHW = imageHWC.transpose((2, 0, 1))
saliency = saliency.flatten()
input_arrays = np.zeros([1, 3, 96, 96], np.float32)
input_arrays[0] = imageCHW