def display_details():
# グラフの描画
if display == True:
dis = np.vstack((dis, np.array([precision, recall, f])))
if (i + 1) % variety == 0:
ax2.plot(dis[:, 0],
dis[:, 1],
"-o",
label='sigma = {}'.format(param[1]),
color='r')
ax2.legend()
dis = np.empty((0, 3), int)
# rangeのscoreも出すとき
if detail == True:
pred_seed_img = rd.main(img, params[best])
precision, recall, true_features, detect_img = ev.main(
pred_seed_img, true_ror, label_img, true_features)
range_p, range_r = get_detail_score(true_features, pred_seed_img)
print 'Prc by range : {}'.format(range_p)
print 'Rcl by range : {}'.format(range_r)
# 検出結果を画像で出すとき
if result_image == True:
result_best = get_detect_img(img, detect_img)
result_stock = get_detect_img(img, detect_stock)
cv2.imshow('best result', result_best)
cv2.imshow('total result', result_stock)
def main(img, true_ror):
'''
args : src -> 3ch, input image
true_ror -> 1ch, ground-truth binary image
'''
## get params
params = get_params('test')
scores = []
pre2s = []
recall2s = []
## True feature
true_features = fe.main(img, true_ror.copy())
true_models = md.main(true_features, N=10)
composition, size, shape = ad.main(true_features, true_models) ; print 'Complete AD'
raw_input()
# ad.main() ; print 'Complete trueAD'
for i, param in enumerate(params):
if i == 1:
'''Target Detection'''
print 'Start detection'
pred_ror, pred_seed = rd.main(img, param=param)
pred_features = fe.main(img, pred_ror)
pred_models = md.main(pred_features, N=10)
# composition, size, shape = ad.main(pred_features, pred_models) ; print 'Complete AD'
'''Evaluate'''
# pre, recall, f, detect_img = seedev.main(pred_seed, true_ror)
# convex, pre2, recall2 = rgev.main(pred_ror, true_ror)
'''Display results'''
# display_3d(pred_features)
display_3d(img, pred_features)
display_models(pred_models[0:4], true_models[0:4], pred_features, true_features)
# display_result(true_ror, param, pred_seed, detect_img, pre, recall, f, i, len(params))
# display_result2(img, detect_img, pred_ror, true_ror, pred_seed) # 画像のスコア
# display_result3(convex, pre2, recall2, param)
# pre2s.append(pre2)
# recall2s.append(recall2)
plt.show()
def main(img, true_ror):
'''
args : src -> 3ch, input image
true_ror -> 1ch, ground-truth binary image
'''
## get params
params = get_params('test')
scores = []
pre2s = []
recall2s = []
## True feature
true_features = fe.main(img, true_ror.copy())
true_models = md.main(true_features, N=10)
composition, size, shape = ad.main(true_features, true_models)
print 'Complete AD'
raw_input()
# ad.main() ; print 'Complete trueAD'
for i, param in enumerate(params):
if i == 1:
'''Target Detection'''
print 'Start detection'
pred_ror, pred_seed = rd.main(img, param=param)
pred_features = fe.main(img, pred_ror)
pred_models = md.main(pred_features, N=10)
# composition, size, shape = ad.main(pred_features, pred_models) ; print 'Complete AD'
'''Evaluate'''
# pre, recall, f, detect_img = seedev.main(pred_seed, true_ror)
# convex, pre2, recall2 = rgev.main(pred_ror, true_ror)
'''Display results'''
# display_3d(pred_features)
display_3d(img, pred_features)
display_models(pred_models[0:4], true_models[0:4], pred_features,
true_features)
# display_result(true_ror, param, pred_seed, detect_img, pre, recall, f, i, len(params))
# display_result2(img, detect_img, pred_ror, true_ror, pred_seed) # 画像のスコア
# display_result3(convex, pre2, recall2, param)
# pre2s.append(pre2)
# recall2s.append(recall2)
plt.show()
def display_details():
# グラフの描画
if display == True:
dis = np.vstack((dis, np.array([precision,recall,f])))
if (i+1)%variety == 0:
ax2.plot(dis[:,0], dis[:,1], "-o",label='sigma = {}'.format(param[1]), color='r')
ax2.legend()
dis = np.empty((0,3), int)
# rangeのscoreも出すとき
if detail == True:
pred_seed_img = rd.main(img, params[best])
precision, recall, true_features, detect_img = ev.main(pred_seed_img, true_ror, label_img, true_features)
range_p, range_r = get_detail_score(true_features, pred_seed_img)
print 'Prc by range : {}'.format(range_p)
print 'Rcl by range : {}'.format(range_r)
# 検出結果を画像で出すとき
if result_image == True:
result_best = get_detect_img(img, detect_img)
result_stock = get_detect_img(img, detect_stock)
cv2.imshow('best result',result_best)
cv2.imshow('total result',result_stock)
def main(src, true_ror):
'''
'''
img = copy.deepcopy(src)
params, variety = get_params()
true_ror, label_img, true_features = get_true_features(img, true_ror)
for i, param in enumerate(params):
## Rock detection and evaluation
pred_ror, pred_seed_img = rd.main(img, param)
precision, recall, true_features, detect_img = ev.main(
pred_seed_img, true_ror, label_img, true_features)
f = 2. * recall * precision / (recall + precision)
print '======================='
print 'Seed threshold : {}'.format(int(param[0]))
print 'Gabor sigma : {}'.format(param[1])
print 'Gabor bias : {}'.format(param[2])
print 'Sun direction : {}π'.format(param[3] / np.pi)
print 'Rock quantity : {}'.format(np.amax(label_img))
print 'Seed quantity : {}'.format(len(np.nonzero(pred_seed_img)[0]))
print 'Correct seeds : {}'.format(true_features[:, 0].sum())
print 'Detected rocks : {}'.format(
(np.array(np.nonzero(true_features[:, 0]))).shape[1])
print 'Precision : {}'.format(round(precision, 3))
print 'Recall : {}'.format(round(recall, 3))
print 'F-measure : {}'.format(round(f, 3))
print 'Ateempt number : {} / {}'.format(i + 1, params.shape[0])
print '======================='
results = np.vstack((results, np.array([precision, recall, f])))
## Initialization and renew
detect_stock_list += true_features[:, 0]
detect_stock[detect_img == 255] = 255
true_features[:, 0] = 0
# グラフの描画
if display == True:
dis = np.vstack((dis, np.array([precision, recall, f])))
if (i + 1) % variety == 0:
ax2.plot(dis[:, 0],
dis[:, 1],
"-o",
label='sigma = {}'.format(param[1]))
ax2.legend()
dis = np.empty((0, 3), int)
# best scoreの表示
best = np.argmax(results[:, 2])
print '***** BEST SCORE ******'
print 'Seed threshold : {}'.format(int(params[best][0]))
print 'Gabor sigma : {}'.format(params[best][1])
print 'Gabor bias : {}'.format(params[best][2])
print 'Sun direction : {}π'.format(params[best][3] / np.pi)
print 'Precision : {}'.format(round(results[best, 0], 3))
print 'Recall : {}'.format(round(results[best, 1], 3))
print 'F-measure : {}'.format(round(results[best, 2], 3))
print 'Ateempt number : {} / {}'.format(best + 1, params.shape[0])
print '***********************'
# rangeのscoreも出すとき
if detail == True:
pred_ror, pred_seed_img = rd.main(img, params[best])
precision, recall, true_features, detect_img = ev.main(
pred_seed_img, true_ror, label_img, true_features)
range_p, range_r = get_detail_score(true_features, pred_seed_img)
print 'Prc by range : {}'.format(range_p)
print 'Rcl by range : {}'.format(range_r)
# 検出結果を画像で出すとき
if result_image == True:
result_best = get_detect_img(img, detect_img)
result_stock = get_detect_img(img, detect_stock)
cv2.imshow('best result', result_best)
cv2.imshow('total result', result_stock)
def main(src, true_ror):
'''
'''
img = copy.deepcopy(src)
params, variety = get_params()
true_ror, label_img, true_features = get_true_features(img, true_ror)
for i, param in enumerate(params):
## Rock detection and evaluation
pred_ror, pred_seed_img = rd.main(img, param)
precision, recall, true_features, detect_img = ev.main(pred_seed_img, true_ror, label_img, true_features)
f = 2.*recall*precision / (recall+precision)
print '======================='
print 'Seed threshold : {}'.format(int(param[0]))
print 'Gabor sigma : {}'.format(param[1])
print 'Gabor bias : {}'.format(param[2])
print 'Sun direction : {}π'.format(param[3]/np.pi)
print 'Rock quantity : {}'.format(np.amax(label_img))
print 'Seed quantity : {}'.format(len(np.nonzero(pred_seed_img)[0]))
print 'Correct seeds : {}'.format(true_features[:,0].sum())
print 'Detected rocks : {}'.format((np.array(np.nonzero(true_features[:,0]))).shape[1])
print 'Precision : {}'.format(round(precision, 3))
print 'Recall : {}'.format(round(recall , 3))
print 'F-measure : {}'.format(round(f , 3))
print 'Ateempt number : {} / {}'.format(i+1,params.shape[0])
print '======================='
results = np.vstack((results, np.array([precision,recall,f])))
## Initialization and renew
detect_stock_list += true_features[:,0]
detect_stock[detect_img == 255] = 255
true_features[:,0] = 0
# グラフの描画
if display == True:
dis = np.vstack((dis, np.array([precision,recall,f])))
if (i+1)%variety == 0:
ax2.plot(dis[:,0], dis[:,1], "-o",label='sigma = {}'.format(param[1]))
ax2.legend()
dis = np.empty((0,3), int)
# best scoreの表示
best = np.argmax(results[:,2])
print '***** BEST SCORE ******'
print 'Seed threshold : {}'.format(int(params[best][0]))
print 'Gabor sigma : {}'.format(params[best][1])
print 'Gabor bias : {}'.format(params[best][2])
print 'Sun direction : {}π'.format(params[best][3]/np.pi)
print 'Precision : {}'.format(round(results[best,0], 3))
print 'Recall : {}'.format(round(results[best,1], 3))
print 'F-measure : {}'.format(round(results[best,2], 3))
print 'Ateempt number : {} / {}'.format(best+1,params.shape[0])
print '***********************'
# rangeのscoreも出すとき
if detail == True:
pred_ror, pred_seed_img = rd.main(img, params[best])
precision, recall, true_features, detect_img = ev.main(pred_seed_img, true_ror, label_img, true_features)
range_p, range_r = get_detail_score(true_features, pred_seed_img)
print 'Prc by range : {}'.format(range_p)
print 'Rcl by range : {}'.format(range_r)
# 検出結果を画像で出すとき
if result_image == True:
result_best = get_detect_img(img, detect_img)
result_stock = get_detect_img(img, detect_stock)
cv2.imshow('best result',result_best)
cv2.imshow('total result',result_stock)