def predictImageClasse():
model = MobileNet()
img = np.asarray(Image.open('./image.jpg'))
img = resize(img, [224, 224])
imshow(img)
X = np.reshape(img, [1, 224, 224, 3])
preds = model.predict(X)
#The 5 most likely between 1000 image net classes
print(tabulate(decode_predictions(preds, top=5)[0], headers=['Name', 'Probability']))
return decode_predictions(preds, top=5)[0]
def predict(self, images, top=None):
if top is None:
top = self.num_classes
images = preprocess_input(images)
predictions = self.model.predict(images)
labels = decode_predictions(predictions, top=top)
return labels
def predict():
# POST
if request.method == 'POST':
# ファイルが読み込まれていない場合は'/predict'に戻る
if 'file' not in request.files:
flash('No file.')
return redirect(url_for('predict'))
# ファイルが読み込まれている場合はそのファイルを読み込む
file = request.files['file']
if file.filename == '':
flash('No file.')
return redirect(url_for('predict'))
# 読み込んだファイルを処理する
if file and is_allowed_file(file.filename):
# 安全なファイル名を作成して画像ファイルを保存
filename = secure_filename(file.filename)
#filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename)
filepath = filename
file.save(filepath)
# 学習済みのMobileNetをロード
# 構造とともに学習済みの重みも読み込まれる
model = MobileNet(weights='imagenet')
# model.summary()
# 引数で指定した画像ファイルを読み込む
# サイズはVGG16のデフォルトである224x224にリサイズされる
img = image.load_img(filepath, target_size=(224, 224))
# 画像ファイルをサーバから削除
os.remove(filepath)
# 読み込んだPIL形式の画像をarrayに変換
x = image.img_to_array(img)
# 3次元テンソル（rows, cols, channels) を
# 4次元テンソル (samples, rows, cols, channels) に変換
# 入力画像は1枚なのでsamples=1でよい
x = np.expand_dims(x, axis=0)
# Top-5のクラスを予測する
# VGG16の1000クラスはdecode_predictions()で文字列に変換される
pred = model.predict(preprocess_input(x))
top = decode_predictions(pred, top=5)[0]
scores = pred[0]
# 予測結果をリストに格納する
results = []
for i in range(5):
score_rounddown = int(top[i][2]*1000000) / 10000.0
results.append([top[i][1], score_rounddown])
return render_template('result.html', results=results)
return render_template('predict.html')
def predict():
# initialize the data dictionary that will be returned from the
# view
data = {"success": False}
# ensure an image was properly uploaded to our endpoint
if flask.request.method == "POST":
if flask.request.files.get("image"):
# read the image in PIL format
image = flask.request.files["image"].read()
image = Image.open(io.BytesIO(image))
# preprocess the image and prepare it for classification
image = prepare_image(image, target=(224, 224))
# classify the input image and then initialize the list
# of predictions to return to the client
preds = model.predict(image)
results = decode_predictions(preds)
data["predictions"] = []
# loop over the results and add them to the list of
# returned predictions
for (imagenetID, label, prob) in results[0]:
r = {"label": label, "probability": float(prob)}
data["predictions"].append(r)
# indicate that the request was a success
data["success"] = True
# return the data dictionary as a JSON response
return flask.jsonify(data)
def label_image(image, model, verbose=False):
# Reference: https://github.com/glouppe/blackbelt/
# convert the image pixels to a numpy array
image = img_to_array(image)
# reshape data for the model
image = np.expand_dims(image, axis=0)
# prepare the image for the VGG model
image = preprocess_input(image)
# predict the probability across all output classes
yhat = model.predict(image)
if verbose and model.output_names[0] == 'predictions':
# convert the probabilities to class labels
labels = decode_predictions(yhat)
# retrieve the most likely result, e.g. highest probability
label = labels[0][0]
# print the classification
print('%s (%.2f%%)' % (label[1], label[2] * 100))
return yhat
def bionic_visual_cortex():
data = {"success": False}
if request.method == 'POST':
if request.files.get('file'):
# read the file
file = request.files['file']
filename = file.filename
filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename)
file.save(filepath)
#format image for processing
image_size = (224, 224)
im = keras.preprocessing.image.load_img(filepath,
target_size=image_size,
grayscale=False)
image = prepare_image(im)
#classify image with accuracy predictions
global graph
with graph.as_default():
preds = model.predict(image)
results = decode_predictions(preds)
print(results)
data["predictions"] = []
for (imagenetID, label, prob) in results[0]:
r = {"label": label, "probability": float(prob)}
data["predictions"].append(r)
data["success"] = True
return jsonify(data)
return '''
def predict(image):
model = MobileNet()
pred = model.predict(image)
decoded_predictions = decode_predictions(pred, top=10)
response = 'MobileNet predictions: ' + str(decoded_predictions[0][0:5])
print(response)
np.argmax(pred[0])
return response
def predict(image):
image = image.resize((224, 224), Image.ANTIALIAS)
image = img_to_array(image)
image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))
image = preprocess_input(image)
yhat = model.predict(image)
label = decode_predictions(yhat)
label = label[0][0]
return label[1].replace("_", " ")
def predict(self, img):
img = image.load_img(img, target_size=(128, 128))
img = image.img_to_array(img).reshape(1, 128, 128, 3)
pred = self.model.predict(preprocess_input(img))
top = decode_predictions(pred, top=5)
desc = []
score = []
for i in range(5):
desc.append(top[0][i][1])
score.append(top[0][i][2] * 100)
return desc, score
def classify(self, url):
original = self.load_resize_img(url)
numpy_image = img_to_array(original)
image_batch = np.expand_dims(numpy_image, axis=0)
processed_image = preprocess_input(image_batch.copy())
labels = []
with self.graph.as_default():
with self.session.as_default():
predictions = self.model.predict(processed_image)
labels = decode_predictions(predictions, top=5)
return labels[0]
def heatmap_for_top_pred(img_path, model, figsizeX):
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
conv_out, preds = model.predict(x)
decode_preds = decode_predictions(preds, top=13)
analysed_preds = 0
top13_preds = preds[0].argsort()[-13:][::-1]
analyzed_class = top13_preds[analysed_preds]
top_class_output = model.output[1][:, analyzed_class]
last_conv_layer = model.get_layer('conv_pw_13')
grads = K.gradients(top_class_output, last_conv_layer.output)[0]
pooled_grads = K.mean(grads, axis=(0, 1, 2))
iterate = K.function([model.input],
[pooled_grads, last_conv_layer.output[0]])
pooled_grads_value, conv_layer_output_value = iterate([x])
for i in range(1024):
conv_layer_output_value[:, :, i] *= pooled_grads_value[i]
heatmap = np.mean(conv_layer_output_value, axis=-1)
heatmap = np.maximum(heatmap, 0)
heatmap /= np.max(heatmap)
heatmap = cv2.resize(heatmap, (224, 224))
heatmap = np.uint8(255 * heatmap)
left, up, down, right = get_bounds(heatmap, percentile=95)
rect = patches.Rectangle((left, up), (right - left), (down - up),
linewidth=1,
edgecolor='r',
facecolor='none')
fig, axes = plt.subplots(nrows=1, ncols=1, figsize=(figsizeX, figsizeX))
axes.imshow(img, alpha=0.7)
axes.imshow(heatmap, cmap='jet', alpha=0.3)
left, up, down, right = get_bounds(heatmap, percentile=95)
rect = patches.Rectangle((left, up), (right - left), (down - up),
linewidth=1,
edgecolor='r',
facecolor='none')
axes.add_patch(rect)
axes.set_title('Heat map and bounding box for prediction: ' +
str(decode_preds[0][analysed_preds][1]))
return None
def getPrediction(filename):
model = MobileNet()
image = load_img('images/' + filename, target_size=(224, 224))
image = img_to_array(image)
image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))
image = preprocess_input(image)
yhat = model.predict(image)
label = decode_predictions(yhat)
label = label[0][0]
print('%s (%.2f%%)' % (label[1], label[2] * 100))
return label[1], label[2] * 100
def mbn(self, name):
img_path = name
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
features = model.predict(x)
preds = decode_predictions(features, top=1)[0]
english_prediction = preds[0][1]
return dictionary[english_prediction]
def run(self):
global not_exist
global camera
global rawCapture
global graph
with graph.as_default():
# allow the camera to warmup
time.sleep(0.1)
# capture frames from camera
for frame in camera.capture_continuous(rawCapture,
format="bgr",
use_video_port=True):
print("read a frame")
# grab the raw NumPy array representing the image, then initialize the timestamp
# and occupied/unoccupied text
image = frame.array
# machine learning
# resize to mobile size(224, 224)
img = Image.fromarray(np.uint8(image))
img = img.resize((224, 224))
x = img
pred_data = np.expand_dims(x, axis=0)
preds = model.predict(preprocess_input(pred_data))
results = decode_predictions(preds, top=1)[0]
item = ''
for result in results:
# print(result)
label = result[1]
item = label
accu = str(result[2])
# gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# show the frame
new_label = label + accu
image = cv2.putText(image, new_label, (10, 25),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0),
2)
cv2.imshow("Frame", image)
key = cv2.waitKey(1) & 0xFF
# 止まるように
if item == stop_item:
not_exist = False
# clear the stream in preparation for the next frame
rawCapture.truncate(0)
# if the `q` key was pressed, break from the loop
if key == ord("q"):
break
def predict(filepath):
#read image
#img = cv2.imread(filepath)
#resize image
#img_resize = cv2.resize(img, (224,224))
img = image.load_img(filepath, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
preds = loaded_model.predict(x)
preds = decode_predictions(preds, top=3)[0]
os.remove(filepath)
return preds
def image_classification(paths):
classification_dict = {}
model = MobileNet(input_shape=None, alpha=1.0, depth_multiplier=1, include_top=True, weights='imagenet', input_tensor=None, pooling=None, classes=1000)
for path in paths:
image = mnetv2_loadimage(path)
preds = model.predict(image)
prediction = decode_predictions(preds, top=1)[0][0]
img_class = prediction[1]
img_confidence = prediction[2]
if img_class in classification_dict:
classification_dict[img_class].append((path, img_confidence))
else:
classification_dict[img_class] = [(path, img_confidence)]
return classification_dict
def mbn(name):
print('requesting MBN...')
dictionary = pickle.load(open('dict.p', 'rb'))
img_path = name
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
features = model.predict(x)
preds = decode_predictions(features, top=1)[0]
english_prediction = preds[0][1]
return dictionary[english_prediction]
def classify(self):
self.write(
"------------------------------------------------------------")
self.write("classify start")
self.write(self.filename)
input = np.stack([self.image])
predictions = self.model.predict(input)
results = decode_predictions(predictions, top=5)[0]
for result in results:
self.write(str(result))
self.write("classify end")
def main(_):
host, port = FLAGS.server.split(':')
channel = implementations.insecure_channel(host, int(port))
stub = prediction_service_pb2.beta_create_PredictionService_stub(channel)
# Send request
#with open(FLAGS.image, 'rb') as f:
# See prediction_service.proto for gRPC request/response details.
img = read_image_as_nparr_RGB(FLAGS.image, shape=(224, 224))
img = preprocess_input(img)
img = np.expand_dims(img, axis=0)
for _ in range(5):
s = time.time()
request = predict_pb2.PredictRequest()
request.model_spec.name = 'mobilenet-classify'
request.model_spec.signature_name = 'predict'
request.inputs['images'].CopyFrom(
tf.make_tensor_proto(img, shape=img.shape)
)
result = stub.Predict(request, 20.0) # 10 secs timeout
#print result.ListFields()
#print tf.contrib.util.make_ndarray(result.ListFields()[0][1].get('scores'))
#print tf.contrib.util.make_ndarray(result.ListFields()[0][1].get('features')).shape
sh = decode_predictions( np.array([result.outputs['features'].float_val]) )
#print(result)
print 'Total Run Time:', time.time() - s, sh
for _ in range(5):
s = time.time()
request = predict_pb2.PredictRequest()
request.model_spec.name = 'mobilenet-alpha-1-228-bottleneck'
request.model_spec.signature_name = 'predict'
request.inputs['images'].CopyFrom(
tf.make_tensor_proto(img, shape=img.shape)
)
result = stub.Predict(request, 20.0) # 10 secs timeout
sh = np.array( result.outputs['features'].float_val ).shape
#print(result)
print 'Total Run Time:', time.time() - s, sh
def upload():
if request.method == 'POST':
# Get the file from post request
f = request.files['file']
# Save the file to ./uploads
basepath = os.path.dirname(__file__)
file_path = os.path.join(basepath, 'uploads',
secure_filename(f.filename))
f.save(file_path)
# Make prediction
preds = model_predict(file_path, model)
pred_class = decode_predictions(preds, top=5) # ImageNet Decode
result = str(pred_class[0][0][1]) # Convert to string
return 'The model MobileNet predicts this image as : ' + result
return None
def graph_reco(path):
model = MobileNet(weights='imagenet')
img_path = path
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
preds = model.predict(x)
predict_list = decode_predictions(preds, top=3)[0]
to_trans = []
for i in predict_list:
to_trans.append(' '.join(str(i[1]).split('_')))
translator = tr.Youdao_translate()
res = []
for i in to_trans:
res.append(translator.get_translation(i))
return res
def img_pred(image):
# 読み込んだ画像を行列に変換
img_array = img_to_array(image)
# 3次元を4次元に変換
img_dims = np.expand_dims(img_array, axis=0)
# Top3のクラスの予測
with graph.as_default():
preds = model.predict(preprocess_input(img_dims))
# imagenet_class_index.json をダウンロードし、その中から認識結果を表示
results = decode_predictions(preds, top=3)[0]
# resultsを整形
result = [
result[1] + " : " + "{:.2%}".format(result[2]) for result in results
]
return result
def upload():
if request.method == 'POST':
# Get the file from post request
f = request.files['image']
# Save the file to ./uploads
basepath = os.path.dirname(__file__)
file_path = os.path.join(
basepath, 'uploads', secure_filename(f.filename))
f.save(file_path)
# Make prediction
preds = model_predict(file_path, model)
# Process your result for human
# pred_class = preds.argmax(axis=-1) # Simple argmax
pred_class = decode_predictions(preds, top=1) # ImageNet Decode
result = str(pred_class[0][0][1]) # Convert to string
return result
return None
def is_dog(img_path):
"""Predicts if image contains a dog
Args:
img_path (str): String containing path to image for inference
Returns:
bool: Returns whether the image contains a dog according to model
"""
K.clear_session()
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
model = MobileNet(weights='imagenet')
preds = model.predict(x)
preds = decode_predictions(preds, top=1)[0][0]
print(preds)
label = int(preds[0][2:])
return 2085620 <= label and label <= 2113978
def camera():
global does_exist
for frame in my_camera.capture_continuous(rawCapture,
format="bgr",
use_video_port=True):
# grab the raw NumPy array representing the image, then initialize the timestamp
# and occupied/unoccupied text
image = frame.array
# machine learning
# resize to mobile net size(224, 224)
img = Image.fromarray(np.uint8(image))
img = img.resize((224, 224))
x = img
pred_data = np.expand_dims(x, axis=0)
with graph.as_default():
preds = model.predict(preprocess_input(pred_data))
results = decode_predictions(preds, top=1)[0]
for result in results:
# print(result)
label = result[1]
accu = str(result[2])
# gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# show the frame
new_label = label + accu
image = cv2.putText(image, new_label, (10, 25),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)
cv2.imshow("Frame", image)
key = cv2.waitKey(1) & 0xFF
if label == stop_item:
does_exist = True
rawCapture.truncate(0)
time.sleep(video_frame_step)
if key == ord("q"):
break
def eval_boxes(input_image,images,coord_ini,coord_fin,N,model,model_input_shape,threshold=0.5,dic=None):
"paso las imágenes cortadas y evaluo prob, si tiene mayor que threshold guardo coordenadas, label y prob"
good_crops=[]
coord=[]
probs=[]
time_to_resize=0
time_to_predict=0
time_to_preprocess=0
time_to_decodepreds=0
time_appending=0
for i in range(N):
time_1=datetime.datetime.now()
imag_rsz = cv2.resize(images[i], (model_input_shape, model_input_shape)) #no hace big copy
time_2=datetime.datetime.now()
to_pred = np.expand_dims(imag_rsz, axis=0)
to_pred = preprocess_input(to_pred)
time_3=datetime.datetime.now()
preds = model.predict(to_pred)
time_4=datetime.datetime.now()
cod,name,max_prob = decode_predictions(preds, top=1)[0][0]
time_5=datetime.datetime.now()
if max_prob>threshold:
good_crops.append(images[i])
coord.append((coord_ini[i],coord_fin[i]))
probs.append((name,max_prob))
time_6=datetime.datetime.now()
time_to_resize+=(time_2-time_1).total_seconds()
time_to_preprocess+=(time_3-time_2).total_seconds()
time_to_predict+=(time_4-time_3).total_seconds()
time_to_decodepreds+=(time_5-time_4).total_seconds()
time_appending+=(time_6-time_5).total_seconds()
# print(len(probs),'shapeprob')
print('time resizing: ',time_to_resize)
print('time preprocesing: ',time_to_preprocess)
print('time predicting: ', time_to_predict)
print('time decoding preds: ', time_to_decodepreds)
print('time appending', time_appending)
draw_boxes(input_image,coord,probs)
return good_crops,coord,probs
def lambda_handler(event, context):
try:
bucket_name = event['Records'][0]['s3']['bucket']['name']
file_key = event['Records'][0]['s3']['object']['key']
print('Reading {} from S3 bucket {}'.format(file_key, bucket_name))
obj = s3.get_object(Bucket=bucket_name, Key=file_key)
s3Img = obj['Body'].read()
img = image.load_img(BytesIO(s3Img), target_size=(224, 224))
img = image.img_to_array(img)
img = np.expand_dims(img, axis=0)
pImg = preprocess_input(img)
preds = model.predict(pImg)
prediction = decode_predictions(preds, top=1)[0][0]
print(prediction)
result = {
"status": "ok",
"name": file_key,
"result": str(prediction[1]),
"confidence": float(prediction[2])
}
snsTopicArn = os.environ["snsTopicArn"]
response = sns.publish(TargetArn=snsTopicArn,
Message=json.dumps(
{'default': json.dumps(result)}),
MessageStructure='json')
print(response)
except Exception as e:
print(e)
raise e
def transform_predictions_to_api_response(predictions, model_name):
""" Transform predictions into readable response
Args:
predictions: prediction object from tensorflow
model_name: name of the model used for this predictions
Returns:
transformed result object as array
"""
decoded_predictions = {}
if model_name == 'inception_v3':
decoded_predictions = inception_v3.decode_predictions(
np.array(predictions['predictions']))[0]
elif model_name == 'mobilenet':
decoded_predictions = mobilenet.decode_predictions(
np.array(predictions['predictions']))[0]
elif model_name == 'mobilenet_v2':
decoded_predictions = mobilenet_v2.decode_predictions(
np.array(predictions['predictions']))[0]
result = []
for prediction in decoded_predictions:
result.append({'label': prediction[1], 'score': prediction[2]})
return result
def _predict_image(self, img_list, top=10):
"""
画像判定モデルを用いて、何が写っているか判定する。
:param img_list: 判定画像のリスト
:param top: 上位何位まで判定するか
:return: 検知結果 (単語ID, 名称, 判定スコア)のタプルがtopで指定された数のリストになる。
"""
# 判定画像を配列化する。
img_array_list = []
for img in img_list:
resized = img.resize((224, 224))
img_array = image.img_to_array(resized)
img_array = preprocess_input(img_array)
img_array_list.append(img_array)
# バッチサイズ分ずつ取り出し、判定にかける。
ret = []
for index in range(0, len(img_array_list), BATCH):
x = np.array(img_array_list[index:index + BATCH])
preds = self._model.predict(x)
decoded = decode_predictions(preds, top=top)
ret.extend(decoded)
return ret
def get_predictions(model, image):
preds = model.predict(image)
dec_preds = decode_predictions(preds)[0]
_, label1, conf1 = decode_predictions(preds)[0][0]
return label1, conf1, dec_preds
rawCapture = PiRGBArray(camera, size=(640, 480))
# allow the camera to warmup
time.sleep(0.1)
for frame in camera.capture_continuous(rawCapture, format="bgr", use_video_port=True):
# frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
img = Image.fromarray(np.uint8(frame))
img = img.resize((224, 224))
img = image.img_to_array(img)
image = np.expand_dims(image, axis=0)
preds = model.predict(preprocess_input(image))
results = decode_predictions(preds, top=1)[0]
for result in results:
# print(result)
label = result[1]
prob = round(result[2] * 100, 2)
frame = cv2.putText(frame, label, (10, 25), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)
# show the frame
cv2.imshow("Frame", image)
key = cv2.waitKey(1) & 0xFF
# clear the stream in preparation for the next frame
rawCapture.truncate(0)