def gera_GABOR_GLCM_LPB_features(sujeito):
lstImagens = input.load_image('./Publication_Dataset/' + sujeito +
'/TIFFs/8bitTIFFs/')
volumeFeatures = []
for imagem in lstImagens:
frame_denoise = input.apply_filter(imagem, 'anisotropic')
bdValue, new = flatten.flat_image(frame_denoise)
crop = cropping.croppy_mona(new, bdValue)
image_features = []
image_features += features_extraction.apply_gabor(crop).tolist()
crop2 = crop.astype(int)
image_features += features_extraction.apply_glcm(crop2)
image_features += features_extraction.apply_lbp(crop).tolist()
volumeFeatures += image_features
# features_extraction.apply_sift(crop)
# lstSIFTFeatures.append(features_extraction.apply_sift(crop))
print('gabor', len(volumeFeatures))
# dictionary = features_extraction.apply_BOW(lstSIFTFeatures)
fileObject = open('./gabor_glcm_lbp_repLine/' + sujeito, 'wb')
if (fileObject != None):
print('salvando...')
pickle.dump(volumeFeatures, fileObject)
fileObject.close
def testarEnquadramento():
lstImagens = input.load_image(
'./Publication_Dataset/AMD6/TIFFs/8bitTIFFs/')
for imagem in lstImagens:
# frame_gauss = input.apply_filter(imagem,'gauss')
frame_dif = input.apply_filter(imagem, 'anisotropic')
bdValue, new = flatten.flat_image(frame_dif)
crop = cropping.croppy_mona(new, bdValue)
def extractFeatures(sujeito):
lstImagens = input.load_image('./base_interpol/' + sujeito)
lstGeoFeatures = []
for imagem in lstImagens:
imagem = np.asarray(imagem)
frame_denoise = input.apply_filter(imagem, 'anisotropic')
bdValue, new = flatten.flat_image(frame_denoise)
crop = cropping.croppy_mona(new, bdValue)
lstGeoFeatures.append(geo.run((crop), [crop.shape[0], crop.shape[1]]))
print('geo', len(lstGeoFeatures), len(lstGeoFeatures[0]))
# dictionary = features_extraction.apply_BOW(lstSIFTFeatures)
fileObject = open('./geo_features/' + sujeito, 'wb')
if (fileObject != None):
print('salvando...')
pickle.dump(lstGeoFeatures, fileObject)
fileObject.close
def geraGABORFeatures(sujeito):
lstImagens = input.load_image('./Publication_Dataset/' + sujeito +
'/TIFFs/8bitTIFFs/')
lstGABORFeatures = []
for imagem in lstImagens:
frame_denoise = input.apply_filter(imagem, 'anisotropic')
bdValue, new = flatten.flat_image(frame_denoise)
crop = cropping.croppy_mona(new, bdValue)
lstGABORFeatures.append(features_extraction.apply_gabor(crop).tolist())
# features_extraction.apply_sift(crop)
# lstSIFTFeatures.append(features_extraction.apply_sift(crop))
print('gabor', len(lstGABORFeatures), len(lstGABORFeatures[0]))
# dictionary = features_extraction.apply_BOW(lstSIFTFeatures)
fileObject = open('./gabor_features/' + sujeito, 'wb')
if (fileObject != None):
print('salvando...')
pickle.dump(lstGABORFeatures, fileObject)
fileObject.close
print('tam:', len(volumeInLine))
featuresGeo.append(volumeInLine)
vetLabelsGeo.append(getClass(vol))
print('Gerando arff file for glcm', len(vetLabelsGeo), len(featuresGeo))
arffGenerator.createArffFile('./geo_features/GEODATASET',
featuresGeo, vetLabelsGeo, 'DME,NORMAL',
len(featuresGeo[0]))
# getBaseFeatures()
# extractFeatures("DME7")
# loadFeatures()
# import math
lstImagens = input.load_image('./base_interpol/DME7')
imagem = lstImagens[2]
# print lstImagens[3][79][1]
# print lstImagens[3][76][0]
# map(lambda x: x if x<255 else 0, imagem)
imagem = np.asarray(imagem)
frame_denoise = input.apply_filter(imagem, 'anisotropic')
bdValue, new = flatten.flat_image(frame_denoise)
crop = cropping.croppy_mona(new, bdValue)
plt.figure()
plt.subplot(121)
plt.imshow(crop, 'gray')
plt.show()
y_shop = shop_network.relu6
sess.run(tf.initialize_all_variables())
#shop_path = '/ais/gobi4/fashion/retrieval/test_gallery.json'
shop_path = '/ais/gobi4/fashion/retrieval/alex_full_test_gallery.json'
img_path = '/ais/gobi4/fashion/data/Cross-domain-Retrieval/'
with open(shop_path, 'w') as jsonfile:
with open(
'/ais/gobi4/fashion/data/Cross-domain-Retrieval/list_test_pairs.txt',
'rb') as f:
data = f.readlines()
for line in data:
line = line.split()
#print("line[3]:{0}".format(line[3]))
x = input.load_image(img_path + line[1])
#x = input.load_image(img_path+line[2])
x = x.reshape([1, 227, 227, 3])
feed_dict = {x_shop: x, train_mode: False}
y = sess.run([y_shop], feed_dict=feed_dict)
# y, conv1, lrn1, pool1, conv2, lrn2, pool2, conv3, conv4, conv5,pool3, fc6, relu6_ori, fc7, relu7, fc8 = sess.run([y_shop, shop_network.conv1, shop_network.lrn1, shop_network.pool1, shop_network.conv2, shop_network.lrn2, shop_network.pool2, shop_network.conv3, shop_network.conv4, shop_network.conv5, shop_network.pool3, shop_network.fc6, shop_network.relu6, shop_network.fc7, shop_network.relu7, shop_network.fc8], feed_dict=feed_dict)
y = np.asarray(y)
jsondata = {'id': line[2], 'shop_feature': y.tolist()}
jsonfile.write(json.dumps(jsondata) + '\n')
#fc8 = np.asarray(fc8)
#jsondata = {'fc8': fc8.tolist()}
#jsonfile.write(json.dumps(jsondata)+'\n')
#relu7 = np.asarray(relu7)
#jsondata = {'relu7': relu7.tolist()}
street_network.build(rgb=x_street, flag="street", train_mode=train_mode)
y_street = street_network.relu6
sess.run(tf.initialize_all_variables())
street_path = '/ais/gobi4/fashion/retrieval/alex_full_street_features.json'
img_path = '/ais/gobi4/fashion/data/Cross-domain-Retrieval/'
with open(street_path, 'w') as jsonfile:
with open(
'/ais/gobi4/fashion/data/Cross-domain-Retrieval/test_pairs_category.txt',
'rb') as f:
#data = random.sample(f.readlines(), 200)
data = f.readlines()
for line in data:
line = line.split()
# x1 = string.atoi(line[3])
# y1 = string.atoi(line[4])
# x2 = string.atoi(line[5])
# y2 = string.atoi(line[6])
# x = bbox_input.load_image(img_path+line[1], x1, y1, x2, y2)
# x = x.reshape([1, 227, 227, 3])
street_path = line[1]
x = input.load_image(street_path)
feed_dict = {x_street: x, train_mode: False}
y = sess.run([y_street], feed_dict=feed_dict)
y = np.asarray(y)
jsondata = {'id': line[0], 'street_feature': y.tolist()}
jsonfile.write(json.dumps(jsondata) + '\n')
f.close()
jsonfile.close()
