def getPrCurveMergeToOne(
methods=['MY', 'DRFI', 'GMR', 'MEAN'],
num=None,
):
'''
过时的方法 全部读入内存
'''
imgGt = reduce(
lambda x, y: np.r_[x, y],
map(
lambda name: io.imread(IMG_DIR + name[:-3] + 'png').astype(np.bool)
.ravel(), IMG_NAME_LIST[:num]))
print 'imgGt OK!'
resoultDic = dict(zip(methods, [None] * len(methods)))
print resoultDic.keys()
for name in IMG_NAME_LIST[:num]:
dic = getCoarseDic(name, methods, COARSE_DIR)
for k, v in dic.items():
resoultDic[k] = np.r_[
resoultDic[k],
v.ravel()] if resoultDic[k] != None else v.ravel()
print '%d/%d' % (IMG_NAME_LIST.index(name), num), name
precisions, recalls = [], []
for method in methods:
# p, r ,_ = precisionRecallCurve(imgGt,resoultDic[method])
p, r, _ = precision_recall_curve(imgGt, resoultDic[method])
precisions += [p]
recalls += [r]
print method
return precisions, recalls, methods
def grabCutTest():
coarseMethods = ['MY4','QCUT','DRFI']
imgInd = 1
n_segments,compactness = 200,10
imgName = IMG_NAME_LIST[imgInd]
img,imgGt = readImg(imgName)
coarseDic = getCoarseDic(imgName,coarseMethods)
refinedImg = coarseDic['DRFI']
mask = grabCut(img,refinedImg)
imgCut = img*mask[:,:,np.newaxis]
show([img,imgCut])
show([refinedImg,mask])
def integratImgsBy3wayTest():
from algorithm import getSlic,getCoarseDic,integratImgsBy3way,buildMethodDic
labelMap = getSlic(rgbImg,300)
img = rgbImg
m, n = labelMap.shape
coarseMethods = ['GT']
coarseDic = getCoarseDic(imgName,coarseMethods)
# show(coarseDic)
coarseImgs=coarseDic.values()
g()
refinedImgs=map(lambda f:f(img,coarseImgs,labelMap),buildMethodDic.values())
img = integratImgsBy3way(refinedImgs)
show(img)
def getRefindImgsTest():
IMG_DIR = r'E:\3-experiment\SalBenchmark-master\Data\DataSet1\Imgs/'
COARSE_DIR =r'E:\3-experiment\SalBenchmark-master\Data\DataSet1\Saliency/'
IMG_DIR = 'test/'
COARSE_DIR ='test/'
IMG_NAME_LIST = filter(lambda x:x[-3:]=='jpg',listdir(IMG_DIR))
coarseMethods = ['QCUT','DRFI']
coarseMethods = ['GT']
imgInd = 0
n_segments,compactness = 200,10
imgName = IMG_NAME_LIST[imgInd]
img,imgGt = readImg(imgName)
coarseDic = getCoarseDic(imgName,coarseMethods)
#show(coarsesDic)
sumCoarseImg = getSumCoarseImg(coarseDic)
coarseImgs=coarseDic.values()
labelMap = getSlic(img,n_segments,compactness)
rgb = img
img = sk.color.rgb2lab(img)
#show([mark_boundaries(img,labelMap),imgGt])
# 获得4+4维 distance
degreeVectors, Ws = getVectors(img, labelMap)
vectors = getWeightSum(labelMap, degreeVectors, Ws)
vectorsTrains = []
coarseTrains = []
for coarseImg in coarseImgs:
coarseTrain, vectorsTrainTag = getCoarseTrain(coarseImg, labelMap)
vectorsTrains += list(vectors[vectorsTrainTag])
coarseTrains += list(coarseTrain)
elm = getElm(np.array(vectorsTrains), np.array(coarseTrains))
refined = elm.predict(vectors)[:,0]
refinedImg = valueToLabelMap(labelMap,normalizing(refined))
vectorsImg = valueToLabelMap(labelMap,normalizing(vectors.sum(1)))
show([rgb,refinedImg])
show([rgb,vectorsImg])
show(vectorsImg-refinedImg)
loga(vectorsImg-refinedImg)
g()
def my4test():
#if 1:
from algorithm import *
labelMap = getSlic(rgbImg,200)
maxLabel = labelMap.max()+1
m, n = labelMap.shape
coarseMethods = ['MEAN']
coarseDic = getCoarseDic(imgName,coarseMethods)
# show(coarseDic)
sumCoarseImg = getSumCoarseImg(coarseDic)
coarseImgs=coarseDic.values()
img = sk.color.rgb2lab(rgbImg)
# 获得4+4维 distance
degreeVectors, Ws = getVectors(img, labelMap)
weightSumVectors = getWeightSum(labelMap, degreeVectors, Ws)
diffEdges,diffNeighbors = getAllDiffEdgeAndNeighbor(labelMap,Ws)
vectors = np.append(weightSumVectors,diffEdges,1)
vectors = np.append(vectors,diffNeighbors,1)
vectorsTrains = []
coarseTrains = []
for coarseImg in coarseImgs:
coarseTrain, vectorsTrainTag = getCoarseTrain(coarseImg, labelMap)
vectorsTrains += list(vectors[vectorsTrainTag])
coarseTrains += list(coarseTrain)
elm = getElm(np.array(vectorsTrains), np.array(coarseTrains))
refined = elm.predict(vectors)[:,0]
refinedImg = valueToLabelMap(labelMap,normalizing(refined))
show(mark_boundaries(sk.color.lab2rgb(img),labelMap))
print diffEdges.shape
show(valueToLabelMap(labelMap,diffEdges.sum(1)))
show(valueToLabelMap(labelMap,diffNeighbors[:,:4].sum(1)))
# show(valueToLabelMap(labelMap,diffNeighbors[:,4:].sum(1)))
# show(valueToLabelMap(labelMap,diffNeighbors.sum(1)))
show(valueToLabelMap(labelMap,vectors[:,4:8].sum(1)))
def getPrCurve(
methods=['DRFI', 'GMR', 'MEAN'],
num=None,
):
'''
返回数据库前num张图片的平均PR曲线
dataDic 结构
img->[imgname]->[method]->auc,pr
method->[method]->auc,pr
'''
num = num if num else len(IMG_NAME_LIST)
dataDic = {'img': {}}
for name in IMG_NAME_LIST[:num]:
print '%d/%d' % (IMG_NAME_LIST.index(name), num), name
imgGt = io.imread(IMG_DIR + name[:-3] + 'png').astype(np.bool).ravel()
coarseDic = getCoarseDic(name, methods, COARSE_DIR)
dataDic['img'][name] = {}
for k, v in coarseDic.items():
dataDic['img'][name][k] = {}
_dic = dataDic['img'][name][k]
coarseImg = v.ravel()
p, r, _ = precisionRecallCurve(imgGt, coarseImg)
_dic['pr'] = (p, r)
_dic['auc'] = metrics.roc_auc_score(imgGt.ravel(), coarseImg)
dataDic['method'] = dict(zip(methods, [{} for i in range(len(methods))]))
precisions, recalls = [], []
for method in methods:
_dic = dataDic['method'][method]
p, r = sum([v[method]['pr'][0] for k,v in dataDic['img'].items()])/num,\
sum([v[method]['pr'][1] for k,v in dataDic['img'].items()])/num
_dic['pr'] = (p, r)
_dic['auc'] = metrics.auc(r, p)
precisions += [p]
recalls += [r]
return precisions, recalls, methods, dataDic
def saveImgData(imgName, methods):
'''
save data to LABEL_DATA_DIR
return default method`s AUC
'''
if isinstance(imgName, int):
imgName = IMG_NAME_LIST[imgName]
imgGt = io.imread(IMG_DIR + imgName[:-3] + 'png') / 255.
coarseDic = getCoarseDic(imgName, methods, COARSE_DIR)
data = {}
data['name'] = imgName
data['method'] = {}
for method, refindImg in coarseDic.items():
data['method'][method] = getAucAndPr(imgGt, refindImg, method)
dataName = LABEL_DATA_DIR + imgName[:-3] + 'data'
AUC_METHOD = "DRFI"
data['aucMethod'] = AUC_METHOD
data['auc'] = data['method'][AUC_METHOD][0]
saveData(data, dataName)
return data['auc']
def aucBug():
# from saliency import *
# from algorithm import *
# from analysis import *
ll = [('184.jpg', 0.83881506713016329), ('138.jpg', 0.84727243912368833),
('122.jpg', 0.88417034903928537), ('153.jpg', 0.92387361400695434)]
name = ll[0][0]
imgGt = io.imread(IMG_DIR + name[:-3] + 'png').astype(np.bool) #.ravel()
methods = ['DISC2', 'DRFI', 'GMR', 'MEAN']
method = 'DISC2'
coarseDic = getCoarseDic(name, methods, COARSE_DIR)
print coarseDic.keys()
# show(coarseDic.values())
resoult = coarseDic[method]
show([resoult])
show(imgGt)
##imgGt.ravel()
#resoult.ravel()
p, r, _ = precisionRecallCurve(imgGt, resoult)
#l = zip(resoult,imgGt)
print method, 'auc', metrics.roc_auc_score(imgGt.ravel(), resoult.ravel())
plt.plot(p, r)
plt.show()