def showRegions(h5File,
labelFile,
imgFolder='../../Data/all38044JPG/',
imgType='.jpg'):
[imgs, types] = parseNL(labelFile)
fr = h5py.File(h5File, 'r')
for n in range(0, len(imgs)):
name = imgs[n]
label = types[n]
print name, label
im = imread(imgFolder + name + imgType)
if len(im.shape) == 2:
img = skimage.color.gray2rgb(im)
mask = np.array(fr.get(name))
print set(mask.flatten())
type_num = 6
colors = np.random.randint(0, 255, (type_num, 3))
color_regions = colors[mask]
alpha = 0.6
result = (color_regions * alpha + img * (1. - alpha)).astype(np.uint8)
plt.figure(11)
plt.imshow(im, cmap='gray')
plt.figure(111)
plt.imshow(result)
plt.show()
def calLBPFeaSet(dataFolder,
labelFile,
classNum,
imgType,
gridSize,
sizeRange,
classLabel,
saveName,
imResize=None,
withIntensity=None,
diffResolution=True):
LBPFeaDim = 10 + 18 + 26
if withIntensity:
LBPFeaDim += 3
posParaNum = 4
names, labels = plf.parseNL(labelFile)
if classNum == 4:
auroraData = plf.arrangeToClasses(names, labels, classNum, classLabel)
else:
auroraData, _ = plf.arrangeToClasses(names, labels, classNum,
classLabel)
f = h5py.File(saveName, 'w')
f.attrs['dataFolder'] = dataFolder
ad = f.create_group('auroraData')
for c, imgs in auroraData.iteritems():
ascii_imgs = [n.encode('ascii', 'ignore') for n in imgs]
ad.create_dataset(c, (len(ascii_imgs), ), 'S10', ascii_imgs)
feaSet = f.create_group('feaSet')
posSet = f.create_group('posSet')
for c, imgs in auroraData.iteritems():
feaArr = np.empty((0, LBPFeaDim))
posArr = np.empty((0, posParaNum))
for name in imgs:
imgFile = dataFolder + name + imgType
if imResize:
feaVec, posVec = calImgLBPFeatures(
imgFile,
gridSize,
sizeRange,
imResize=imResize,
withIntensity=withIntensity,
diffResolution=diffResolution)
else:
feaVec, posVec = calImgLBPFeatures(
imgFile,
gridSize,
sizeRange,
withIntensity=withIntensity,
diffResolution=diffResolution)
feaArr = np.append(feaArr, feaVec, axis=0)
posArr = np.append(posArr, posVec, axis=0)
feaSet.create_dataset(c, feaArr.shape, 'f', feaArr)
posSet.create_dataset(c, posArr.shape, 'i', posArr)
f.close()
print saveName + ' saved'
return 0
def generateSpecialCommonBbox(labelFile, savePath, dataFolder, imgType, paras):
names, labels = plf.parseNL(labelFile)
fs = h5py.File(savePath, 'w')
for i in range(len(names)):
name = names[i]
label = labels[i]
imgFile = dataFolder + name + imgType
paras['imgFile'] = imgFile
im = skimage.io.imread(imgFile)
if len(im.shape) == 2:
img = skimage.color.gray2rgb(im)
paras['im'] = img
paras['img'] = img
paras['specialType'] = int(
label) - 1 # 0: arc, 1: drapery, 2: radial, 3: hot-spot
paras['th'] = 0.45
paras['returnRegionLabels'] = [1, 2] # 0: special, 1: rest, 2: common
regions_special, angle_special = special_common_local_proposal(paras)
regions_special = np.array(regions_special)
paras['th'] = 0.25
paras['returnRegionLabels'] = [0, 1] # 0: special, 1: rest, 2: common
regions_common, angle_common = special_common_local_proposal(paras)
regions_common = np.array(regions_common)
labels_special = np.zeros((regions_special.shape[0], ), dtype='i')
labels_special.fill(label)
labels_common = np.zeros((regions_common.shape[0], ), dtype='i')
group = fs.create_group(str(i))
group.attrs['imgFile'] = imgFile
group.attrs['imgName'] = name
d_special = group.create_dataset('bbox_special',
shape=regions_special.shape,
dtype='i',
data=regions_special)
d_common = group.create_dataset('bbox_common',
shape=regions_common.shape,
dtype='i',
data=regions_common)
group.create_dataset('labels_special',
shape=labels_special.shape,
dtype='i',
data=labels_special)
group.create_dataset('labels_common',
shape=labels_common.shape,
dtype='i',
data=labels_common)
d_special.attrs['angle'] = angle_special
d_common.attrs['angle'] = angle_common
print name, 'bbox saved'
fs.close()
return 0
def calFusionFeaSet(dataFolder, labelFile, classNum, feaTypes, imgType,
gridSize, sizeRange, classLabel, saveName):
posParaNum = 4
names, labels = plf.parseNL(labelFile)
if classNum == 4:
auroraData = plf.arrangeToClasses(names, labels, classNum, classLabel)
else:
auroraData, _ = plf.arrangeToClasses(names, labels, classNum,
classLabel)
f = h5py.File(saveName, 'w')
f.attrs['dataFolder'] = dataFolder
ad = f.create_group('auroraData')
for c, imgs in auroraData.iteritems():
ascii_imgs = [n.encode('ascii', 'ignore') for n in imgs]
ad.create_dataset(c, (len(ascii_imgs), ), 'S10', ascii_imgs)
feaSet = f.create_group('feaSet')
posSet = f.create_group('posSet')
for c, imgs in auroraData.iteritems():
feaArr = None
posArr = None
for name in imgs:
imgFile = dataFolder + name + imgType
print imgFile, c
feaVec, posVec = calFusionFea(imgFile, feaTypes, gridSize,
sizeRange)
if feaArr is None:
feaArr = np.empty((0, feaVec.shape[1]))
posArr = np.empty((0, posParaNum))
feaArr = np.append(feaArr, feaVec, axis=0)
posArr = np.append(posArr, posVec, axis=0)
feaSet.create_dataset(c, feaArr.shape, 'f', feaArr)
posSet.create_dataset(c, posArr.shape, 'i', posArr)
f.close()
print saveName + ' saved'
return 0
import numpy as np
import h5py
from src.util.paseLabeledFile import parseNL
if __name__ == '__main__':
labeledFile = '../../Data/type4_b500.txt'
names, labels = parseNL(labeledFile)
bboxFile = '../../Data/type4_b500_bbox.hdf5'
save_bboxFile = '../../Data/type4_b500_whole_bbox.hdf5'
f = h5py.File(bboxFile, 'r')
fs = h5py.File(save_bboxFile, 'w')
for i in range(len(f)):
im_i = f.get(str(i))
imgFile = im_i.attrs['imgFile']
name = im_i.attrs['imgName']
regions_special = im_i.get('bbox_special')
regions_common = im_i.get('bbox_common')
labels_special = im_i.get('labels_special')
labels_common = im_i.get('labels_common')
angle_special = regions_special.attrs['angle']
angle_common = regions_common.attrs['angle']
label = int(labels[names.index(name)])
print regions_special.shape
whole = np.array([[0, 0, 440, 440]])
if regions_special.shape[0] == 0:
regions_special = whole
else:
regions_special = np.append(regions_special, whole, axis=0)
# print label
if labels_special.shape[0] != 0:
import src.preprocess.esg as esg
from scipy.misc import imsave
from src.local_feature.adaptiveThreshold import calculateThreshold
from scipy.misc import imread, imresize
import os
import scipy.io as sio
from src.experiments.test_segmentation_accuracy import load_mask_mat
from src.localization.generateSubRegions import detect_regions
if __name__ == '__main__':
data_folder = '/home/amax/NiuChuang/KLSA-auroral-images/Data/segmentation_data_v2/'
heatmap_folder = '/home/amax/NiuChuang/CAM/aurora_heatmaps/'
predict_file = '/home/amax/NiuChuang/CAM/aurora_predict.txt'
type_num = 4
[names, predicts] = plf.parseNL(predict_file)
k = 60
minSize = 50
sigma = 0.5
imSize = 440
eraseMap = np.zeros((imSize, imSize))
radius = imSize / 2
centers = np.array([219.5, 219.5])
for i in range(imSize):
for j in range(imSize):
if np.linalg.norm(np.array([i, j]) - centers) > radius + 5:
eraseMap[i, j] = 1
cls_confusion = np.zeros((type_num, type_num))
yl = [adaptiveLinear(x) for x in xl]
return yl
def calculateThreshold(imgFile):
hist = calculateImgHist(imgFile)
thresh = adaptiveLinear(hist[180:].sum())
return thresh
if __name__ == '__main__':
imgFile = '/home/niuchuang/PycharmProjects/KLSA-auroral-images/Data/labeled2003_38044/N20031225G100901.bmp'
thresholdLabelFile = '../../Data/threshold_annotations.txt'
imagesFolder_thresh = '../../Data/images_for_threshold/'
imgType = '.jpg'
[names, labels] = plf.parseNL(thresholdLabelFile)
bright_nums = np.zeros((len(names), ))
thresholds = np.zeros((len(names), ))
for i in xrange(len(names)):
imgFile = imagesFolder_thresh + names[i] + imgType
hist = calculateImgHist(imgFile)
bright_nums[i] = hist[150:].sum()
thresholds[i] = int(labels[i])
plt.figure(1)
xs = range(60000)
ys = calAdapList(xs)
plt.plot(xs, ys)
plt.scatter(bright_nums, thresholds, 50, color='blue')
plt.show()
# outCircleSize = int(round(((440*440) - (math.pi * 221 * 221))))
# im = imread(imgFile)
import numpy as np
from src.util.paseLabeledFile import parseNL
from scipy.misc import imread
if __name__ == '__main__':
labelFile = '../../Data/Alllabel2003_38044.txt'
imgsFolder = '../../Data/all38044JPG/'
imgType = '.jpg'
[names, labels] = parseNL(labelFile)
img_num = len(names)
mean = 0
for i in range(img_num):
name = names[i]
imgFile = imgsFolder + name + imgType
im = imread(imgFile)
mean += im.mean()
mean = mean / img_num
print mean
"""According to some self-define rules, this function selects the images for
following experiments."""
import src.util.paseLabeledFile as plf
filePath = '../../Data/Alllabel2003_38044.txt'
file_selected = '../../Data/one_in_minute.txt'
file_balance = '../../Data/balanceSampleFrom_one_in_minute.txt'
file_own = '../../Data/one_in_minute_selected.txt'
class_num = 4
select_num_perclass = 500
[names, labels] = plf.parseNL(filePath)
arrangedImgs = plf.arrangeToClasses(names, labels, class_num)
print 'total labeled number:'
for i in range(4):
print 'NO. class ' + str(i + 1), len(arrangedImgs[str(i + 1)])
# print plf.timeDiff(names[0], names[2])
[ids, sampledImgs] = plf.sampleImages(names)
f_w = open(file_selected, 'w+')
for i in range(len(ids)):
f_w.write(sampledImgs[i] + ' ' + labels[ids[i]] + '\n')
f_w.close()
[names_s, labels_s] = plf.parseNL(file_selected)
patch_mean = float(f_mean.readline().split(' ')[1])
f_mean.close()
# diff mean
# patch_mean = 0
print 'patch_mean: ' + str(patch_mean)
# saveSDAEFeas = '../../Data/Features/type3_SDAEFeas.hdf5'
# saveSDAEFeas = '../../Data/Features/type4_SDAEFeas.hdf5'
# saveSDAEFeas = '../../Data/Features/type4_SDAEFeas_diff_mean.hdf5'
# saveSDAEFeas = '../../Data/Features/type4_SDAEFeas_same_mean_s16_600_300_300_300.hdf5'
# saveSDAEFeas = '../../Data/Features/type4_SDAEFeas_same_mean_s28_b500_special.hdf5'
saveSDAEFeas = '../../Data/Features/type4_SDAEFeas_same_mean_s28_b500_special_trained.hdf5'
# saveSDAEFeas = '../../Data/Features/type4_SDAEFeas_same_mean_s28_b500_special_classification.hdf5'
[images, labels] = plf.parseNL(labelFile)
# arrImgs, _ = plf.arrangeToClasses(images, labels, classNum, classes)
arrImgs = plf.arrangeToClasses(images, labels, classNum, classes)
for i in arrImgs:
print i, len(arrImgs[i])
f = h5py.File(saveSDAEFeas, 'w')
f.attrs['dataFolder'] = dataFolder
ad = f.create_group('auroraData')
for c, imgs in arrImgs.iteritems():
ascii_imgs = [n.encode('ascii', 'ignore') for n in imgs]
ad.create_dataset(c, (len(ascii_imgs), ), 'S10', ascii_imgs)
feaSet = f.create_group('feaSet')
posSet = f.create_group('posSet')
def generateRegions():
paras = {}
paras['k'] = 60
paras['minSize'] = 50
paras['patchSize'] = np.array([16, 16])
paras['region_patch_ratio'] = 0.1
paras['sigma'] = 0.5
paras['alpha'] = 0.6
paras['types'] = ['arc', 'drapery', 'radial', 'hot_spot']
paras[
'lbp_wordsFile_s1'] = '../../Data/Features/type4_LBPWords_s1_s16_b300_w500.hdf5'
paras[
'lbp_wordsFile_s2'] = '../../Data/Features/type4_LBPWords_s2_s16_b300_w500.hdf5'
paras[
'lbp_wordsFile_s3'] = '../../Data/Features/type4_LBPWords_s3_s16_b300_w500.hdf5'
paras[
'lbp_wordsFile_s4'] = '../../Data/Features/type4_LBPWords_s4_s16_b300_w500.hdf5'
eraseMapPath = '../../Data/eraseMap.bmp'
if not os.path.exists(eraseMapPath):
imSize = 440
eraseMap = np.zeros((imSize, imSize))
radius = imSize / 2
centers = np.array([219.5, 219.5])
for i in range(440):
for j in range(440):
if np.linalg.norm(np.array([i, j]) - centers) > 220 + 5:
eraseMap[i, j] = 1
imsave(eraseMapPath, eraseMap)
else:
eraseMap = imread(eraseMapPath) / 255
paras['eraseMap'] = eraseMap
paras['sizeRange'] = (16, 16)
paras['imResize'] = (256, 256)
paras['imgSize'] = (440, 440)
paras['nk'] = 1
resolution = 1
gridSize = np.array([resolution, resolution])
paras['resolution'] = resolution
paras['gridSize'] = gridSize
paras['sdaePara'] = None
paras['feaType'] = 'LBP'
paras['mk'] = 0
paras['withIntensity'] = False
paras['diffResolution'] = False
paras['isSave'] = False
paras['is_rotate'] = False
paras['train'] = False
labelFile = '../../Data/type4_b500.txt'
imageFolder = '../../Data/all38044JPG/'
imgType = '.jpg'
[names, labels] = parseNL(labelFile)
f = h5py.File('../../Data/regions.hdf5', 'w')
# imgFile = '../../Data/labeled2003_38044/N20031221G071131.bmp'
# label = 2
for i in range(len(names)):
name = names[i]
imgFile = imageFolder + name + imgType
label = labels[i]
paras['imgFile'] = imgFile
im = np.array(imread(imgFile), dtype='f') / 255
paras['im'] = im
im1 = skimage.io.imread(imgFile)
if len(im1.shape) == 2:
img = skimage.color.gray2rgb(im1)
paras['img'] = img
paras['thresh'] = calculateThreshold(imgFile)
paras['specialType'] = int(
label) - 1 # 0: arc, 1: drapery, 2: radial, 3: hot-spot
paras['returnRegionLabels'] = [0] # 0: special, 1: rest, 2: common
paras['th'] = 0.45
F0, region_labels, eraseLabels = region_special_map(paras)
# save type: 0: common, 1: arc, 2, drapery, 3: radial, 4: hot-spot, 5: rest
mask = np.zeros(F0.shape, dtype='i')
mask.fill(5)
for i in region_labels:
mask[np.where(F0 == i)] = int(label)
paras['specialType'] = int(
label) - 1 # 0: arc, 1: drapery, 2: radial, 3: hot-spot
paras['returnRegionLabels'] = [2] # 0: special, 1: rest, 2: common
paras['th'] = 0.25
# paras['thresh'] = 0
F0, region_labels, eraseLabels, filterout_labels = region_special_map(
paras, returnFilteroutLabels=True)
# print filterout_labels
region_labels = region_labels + filterout_labels
for i in region_labels:
mask[np.where(F0 == i)] = 0
f.create_dataset(name, mask.shape, dtype='i', data=mask)
print name + ' saved!'
f.close()
return 0
def makePatchData(labelFile,
patchSize,
gridSize=np.array([10, 10]),
imgType='.bmp',
channels=1,
savePath='../../Data/one_in_minute_patch_diff_mean.hdf5',
same_mean_file='../.../Data/patchData_mean_s16.txt',
imagesFolder='../../Data/labeled2003_38044/',
patchMean=True,
saveList='../../Data/patchList_diff_mean.txt',
subtract_same_mean=False):
sizeRange = (patchSize, patchSize)
[images, labels] = plf.parseNL(labelFile)
# arragedImages = plf.arrangeToClasses(images, labels, classNum)
f = h5py.File(savePath, 'w')
data = f.create_dataset('data', (0, channels, patchSize, patchSize),
dtype='f',
maxshape=(None, channels, patchSize, patchSize))
label = f.create_dataset('label', (0, ), dtype='i', maxshape=(None, ))
if subtract_same_mean:
patches_mean = 0
for i in range(len(images)):
imf = imagesFolder + images[i] + imgType
gridPatchData, gridList, _ = esg.generateGridPatchData(
imf, gridSize, sizeRange)
patchData = np.array(gridPatchData)
patches_mean += patchData.mean()
patch_mean = patches_mean / len(images)
print 'patch number: ' + str(data.shape[0])
print 'patch mean: ' + str(patch_mean)
with open(same_mean_file, 'w') as f2:
f2.write('patch_mean: ' + str(patch_mean))
f2.close()
else:
patch_mean = 0
print 'patch_mean: ', patch_mean
for i in range(len(images)):
imf = imagesFolder + images[i] + imgType
print imf
gridPatchData, gridList, _ = esg.generateGridPatchData(
imf, gridSize, sizeRange)
patchData = [
p.reshape(channels, patchSize, patchSize) for p in gridPatchData
]
patchData = np.array(patchData) - patch_mean
if patchMean:
means = np.mean(np.mean(patchData, axis=-1), axis=-1)
means = means.reshape(means.shape[0], means.shape[1], 1, 1)
means = np.tile(means, (1, 1, patchSize, patchSize))
patchData -= means
labelData = np.full((len(gridList), ), int(labels[i]), dtype='i')
oldNum = data.shape[0]
newNum = oldNum + patchData.shape[0]
data.resize(newNum, axis=0)
data[oldNum:newNum, :, :, :] = patchData
label.resize(newNum, axis=0)
label[oldNum:newNum, ] = labelData
f.close()
print 'make patch data done!'
with open(saveList, 'w') as f1:
f1.write(savePath)
f1.close()
print saveList + ' saved!'
return 0
def calCascadeFeaSet(dataFolder,
labelFile,
siftFeaFile_reduce,
lbpFeaFile_reduce,
sdaeFeaFile_reduce_d,
sdaeFeaFile_reduce_s,
classNum,
imgType,
gridSize,
sizeRange,
classLabel,
sdaePara,
saveName,
saveFolder='../../Data/Features/'):
sift_f = h5py.File(siftFeaFile_reduce, 'r')
sdae_f_d = h5py.File(sdaeFeaFile_reduce_d, 'r')
sdae_f_s = h5py.File(sdaeFeaFile_reduce_s, 'r')
lbp_f = h5py.File(lbpFeaFile_reduce, 'r')
names, labels = plf.parseNL(labelFile)
if classNum == 4:
auroraData = plf.arrangeToClasses(names, labels, classNum, classLabel)
else:
auroraData, _ = plf.arrangeToClasses(names, labels, classNum,
classLabel)
f = h5py.File(saveFolder + saveName, 'w')
f.attrs['dataFolder'] = dataFolder
ad = f.create_group('auroraData')
for c, imgs in auroraData.iteritems():
ascii_imgs = [n.encode('ascii', 'ignore') for n in imgs]
ad.create_dataset(c, (len(ascii_imgs), ), 'S10', ascii_imgs)
feaSet = f.create_group('feaSet')
posSet = f.create_group('posSet')
for c, imgs in auroraData.iteritems():
# sift_u = np.array(sift_f.get('uSet/'+c))
# lbp_u = np.array(lbp_f.get('uSet/'+c))
# sdae_u_d = np.array(sdae_f_d.get('uSet/'+c))
# sdae_u_s = np.array(sdae_f_s.get('uSet/'+c))
sift_u = np.array(sift_f.get('uSet/u'))
lbp_u = np.array(lbp_f.get('uSet/u'))
sdae_u_d = np.array(sdae_f_d.get('uSet/u'))
sdae_u_s = np.array(sdae_f_s.get('uSet/u'))
imgFile = dataFolder + imgs[0] + imgType
_, gl, _ = esg.generateGridPatchData(imgFile, gridSize, sizeRange)
feaVec, posVec = extractCascadeFeatures(imgFile, sift_u, lbp_u,
sdae_u_d, sdae_u_s, gl,
gridSize, sizeRange, sdaePara)
feaArr = np.empty((0, feaVec.shape[1]))
posArr = np.empty((0, posVec.shape[1]))
for name in imgs:
imgFile = dataFolder + name + imgType
batchSize = len(gl)
inputShape = (batchSize, 1, sizeRange[0], sizeRange[0])
sdaePara['inputShape'] = inputShape
feaVec, posVec = extractCascadeFeatures(imgFile, sift_u, lbp_u,
sdae_u_d, sdae_u_s, gl,
gridSize, sizeRange,
sdaePara)
feaArr = np.append(feaArr, feaVec, axis=0)
posArr = np.append(posArr, posVec, axis=0)
feaSet.create_dataset(c, feaArr.shape, 'f', feaArr)
posSet.create_dataset(c, posArr.shape, 'i', posArr)
f.close()
print saveFolder + saveName + ' saved'
return 0
