def regionSetToBoxes(regionsSet,
overlapThresh,
sizeRange=[0, 440 * 440],
isVisualize=False):
bboxList = []
for region in regionsSet:
for region_i in region[0]:
r = region_i[1]
y = r[0]
x = r[1]
h = r[2] - y
w = r[3] - x
if (w * h < sizeRange[1]) and (w * h > sizeRange[0]):
bboxList.append([y, x, h, w])
if isVisualize:
regionMap = region[1]
regionLabels = region_i[2]
pseudoMap = np.zeros((440, 440))
for label in regionLabels:
pseudoMap[np.where(regionMap == label)] = 255
plf.showGrid(pseudoMap, [[y, x, h, w]])
plt.show()
bboxes = np.array(bboxList)
keepBoxes = filterOverlap(bboxes, overlapThresh)
keepBoxes = np.array(keepBoxes)
# convert [y1, x1, h, w] to [y1, x1, y2, x2]
keepBoxes[:, 2] = keepBoxes[:, 0] + keepBoxes[:, 2]
keepBoxes[:, 3] = keepBoxes[:, 1] + keepBoxes[:, 3]
# convert [y1, x1, y2, x2] to [x1, y1, x2, y2]
keepBoxes[:, [0, 1]] = keepBoxes[:, [1, 0]]
keepBoxes[:, [2, 3]] = keepBoxes[:, [3, 2]]
return keepBoxes
def show_region_patch_grid(imgFile,
F0,
region_patch_list,
alpha,
eraseMap,
saveFolder=None):
img = skimage.io.imread(imgFile)
if len(img.shape) == 2:
img = skimage.color.gray2rgb(img)
n_region = len(set(list(F0.flatten())))
eraseLabels = set(list(F0[numpy.where(eraseMap == 1)].flatten()))
colors = numpy.random.randint(0, 255, (n_region, 3))
for e in eraseLabels:
colors[e] = 0
color_regions = colors[F0]
result = (color_regions * alpha + img * (1. - alpha)).astype(numpy.uint8)
if saveFolder is not None:
imName = imgFile[-20:-4]
ii = 1
for l in region_patch_list:
if len(l) != 0:
plf.showGrid(result, l)
if saveFolder is not None:
plt.savefig(saveFolder + imName + '_subregion' + str(ii) +
'.jpg')
ii += 1
# plt.imshow(result)
plt.show()
#
# l_s = np.array(labels_special).reshape(b_s.shape[0], 1)
# l_c = np.array(labels_common).reshape(b_c.shape[0], 1)
# print l_s.shape
#
# bb = np.zeros((0, 4), dtype='i')
# ll = np.zeros((0, 1), dtype='i')
# if b_s.shape[0] != 0:
# bb = np.vstack([bb, b_s])
# ll = np.vstack([ll, l_s])
# if b_c.shape[0] != 0:
# ll = np.vstack([ll, l_c])
# bb = np.vstack([bb, b_c])
# print b_s
# print b_c
# print bb
# print l_s
# print l_c
# print ll
im = imread(imgFile)
im_s = rotate(im, angle_special)
bbox_s = list(bbox_special)
bbox_c = list(bbox_common)
plf.showGrid(im_s, bbox_s)
print imgName
plt.title(imgName + '_special')
im_c = rotate(im, angle_common)
plf.showGrid(im_c, bbox_c)
plt.title(imgName + '_common')
plt.show()
paras['imgFile'] = imgFile
im = skimage.io.imread(imgFile)
if len(im.shape) == 2:
img = skimage.color.gray2rgb(im)
paras['im'] = img
paras['img'] = img
save_hierachecalRegionsProcess(paras)
if paras['is_rotate']:
regions, angle = special_common_local_proposal(paras)
else:
regions = special_common_local_proposal(paras)
# regions = list(regions)
print np.array(regions)
if paras['train']:
for k, v in regions.iteritems():
plf.showGrid(img, v)
plt.title(k)
plt.show()
else:
if paras['is_rotate']:
img = rotate(img, angle)
plf.showGrid(img, regions)
plt.savefig(figSaveFolder + name + '.png')
plt.show()
# im = skimage.io.imread(imgFile)
# if len(im.shape) == 2:
# img = skimage.color.gray2rgb(im)
# (R, F, L, L_regions) = selective_search.hierarchical_segmentation(img, k, feature_masks, eraseMap=None)
# print('result filename: %s_[0000-%04d].png' % (out_prefix, len(F) - 1))
# # print feaVecs.shape, posVecs.shape
# labelVecs = chm.calPatchLabels2(w, feaVecs, k=nk, two_classes=['1', '2'], isH1=True)
# posVecs_f, labelVecs_f = filterPos(posVecs, labelVecs, radius=3, spaceSize=10)
# specialIDs = list(np.argwhere(labelVecs_f == 0)[:, 0])
# specialPos = list(posVecs_f[specialIDs, :])
wordsFile_s = [
lbp_wordsFile_s1, lbp_wordsFile_s2, lbp_wordsFile_s3, lbp_wordsFile_s4
]
specialType = 2
w = wordsFile_s[specialType]
patchData_a, specialPos = selectSpecialPatch(imgFile, w, feaType, gridSize,
sizeRange, nk)
# print labelVecs.shape
print len(specialPos)
print len(patchData_a)
im = imread(imgFile)
patchData, _, _ = esg.generateGridPatchData(im,
gridSize,
sizeRange,
gridList=specialPos)
print len(patchData)
pp = np.array(patchData)
background_mean = pp.mean()
print background_mean
mm = np.mean(np.mean(pp, axis=2), axis=1)
print mm.shape
plf.showGrid(im, specialPos)
plt.show()
specialList = threshHoldFilterPatch(im, gridSize, sizeRange)
plf.showGrid(im, specialList)
plt.show()
plt.plot(radius, sizes)
plt.show()
labelFile = '../../Data/balanceSampleFrom_one_in_minute.txt'
imagesFolder = '../../Data/labeled2003_38044/'
imgType = '.bmp'
gridSize = np.array([10, 10])
sizeRange = (16, 16)
[images, labels] = plf.parseNL(labelFile)
# imgFile = imagesFolder + images[0] + imgType
imName = 'N20031221G094901'
saveFolder = '/home/ljm/NiuChuang/KLSA-auroral-images/Data/Results/sample_grid/'
imgFile = imagesFolder + imName + imgType
# im = Image.open(imgFile)
# im = np.array(im)
# imageSize = np.array(im.shape)
# gridList = generateGrid(imageSize, gridSize, sizeRange)
gridPatchData, gridList, im = generateGridPatchData(
imgFile, gridSize, sizeRange)
# print gridList[0:5]
# print len(gridList)
plf.showGrid(im, gridList)
plt.savefig(saveFolder + imName + '_grid.jpg')
# plt.imsave('/home/ljm/NiuChuang/KLSA-auroral-images/Data/Results/sample_grid/N20040118G050641_grid.jpg')
plt.show()