def tailor(anno, src_path, dst_path):
for box in anno['boxs']:
if box["content"] == "Business_card":
region = (box['x'], box['y'], box['x']+box['w'], box['y']+box['h'])
crop(src_path, dst_path, region)
return
logger.error("unable to find annotation for business card in {0}".format(anno['img']))
raise ValueError
def crop_func():
session_id = request.args.get("session_id")
filename = request.args.get("filename")
path = app.config["UPLOAD_FOLDER"] + "/" + session_id + "/" + filename
data = request.json
X = int(data["X"])
Y = int(data["Y"])
xOffset = int(data["xOffset"])
yOffset = int(data["yOffset"])
crop(path, X, Y, xOffset, yOffset)
return jsonify("OK"), 200
def crop_all_imgs(path, anno):
title = os.path.basename(path)
cropped_path = os.path.join(path, 'cropped')
images = filter(lambda x: x.endswith('.png'), os.listdir(path))
if not os.path.exists(cropped_path):
os.makedirs(cropped_path)
from utils.crop import crop
for frame in images:
img_name = os.path.join(path, frame)
dst_name = os.path.join(cropped_path, frame)
region = anno[frame]['region']
# if os.path.exists(dst_name): # only process unfinished
# continue
try:
crop(img_name, dst_name, region)
except Exception as e:
region_str = ','.join(map(lambda x: str(int((x))), region))
logger.error('error occured when cropping {0} with region {1}'.format(img_name, region_str))
model.add(Flatten())
elif(l.type == "CONNECTED"):
model.add(Dense(l.output_size, weights=[l.weights,l.biases]))
model.add(LeakyReLU(alpha=0.1))
else:
print "Error: Unknown Layer Type"
model.add(Activation('softmax'))
return model
def get_activations(model, layer, X_batch):
get_activations = theano.function([model.layers[0].input], model.layers[layer].get_output(train=False), allow_input_downcast=True)
activations = get_activations(X_batch) # same result as above
return activations
#image = readImg(os.path.join(os.getcwd(),'images/dog.file'))
image = crop(os.path.join(os.getcwd(),'images/dog.jpg'))
image = np.expand_dims(image, axis=0)
googleNet = ReadGoogleNetWeights(os.path.join(os.getcwd(),'weights/extraction.weights'))
#reshape weights in every layer
for i in range(googleNet.layer_number):
l = googleNet.layers[i]
if(l.type == 'CONVOLUTIONAL'):
weight_array = l.weights
n = weight_array.shape[0]
weight_array = weight_array.reshape((n//(l.size*l.size),(l.size*l.size)))[:,::-1].reshape((n,))
weight_array = np.reshape(weight_array,[l.n,l.c,l.size,l.size])
l.weights = weight_array
if(l.type == 'CONNECTED'):
weight_array = l.weights
weight_array = np.reshape(weight_array,[l.output_size,l.input_size])
predictions = out[0]
boxes = convert_yolo_detections(predictions)
boxes = do_nms_sort(boxes,98)
for i in range(98):
for j in range(20):
if(boxes[i].probs[j] != 0):
print i,j
print boxes[i].probs[j]
draw_detections(os.path.join(os.getcwd(),'images/dog.jpg'),98,0.2,boxes,20,labels,'dog.jpg')
'''
#for each image, we generate a detection result
imagePath = os.path.join(os.getcwd(),'images')
images = [f for f in listdir(imagePath) if isfile(join(imagePath, f))]
for image_name in images:
timer = Timer()
image = crop(os.path.join(imagePath,image_name),resize_width=512,resize_height=512,new_width=448,new_height=448)
image = np.expand_dims(image, axis=0)
timer.tic()
out = model.predict(image)
timer.toc()
print ('Total time is {:.3f}s ').format(timer.total_time)
predictions = out[0]
boxes = convert_yolo_detections(predictions)
boxes = do_nms_sort(boxes,98)
draw_detections(os.path.join(imagePath,image_name),98,0.2,boxes,20,labels,image_name)
#draw_detections(os.path.join(os.getcwd(),'resized_images','1.jpg'),98,0.2,boxes,20,labels,image_name)
elif(l.type == "CONNECTED"):
model.add(Dense(l.output_size, weights=[l.weights,l.biases]))
model.add(LeakyReLU(alpha=0.1))
elif(l.type == "SOFTMAX"):
model.add(Activation('softmax'))
else:
print "Error: Unknown Layer Type",l.type
return model
def get_activations(model, layer, X_batch):
get_activations = theano.function([model.layers[0].input], model.layers[layer].get_output(train=False), allow_input_downcast=True)
activations = get_activations(X_batch) # same result as above
return activations
#image = readImg(os.path.join(os.getcwd(),'images/dog.file'))
image = crop(os.path.join(os.getcwd(),'images/eagle.jpg'))
image = np.expand_dims(image, axis=0)
darkNet = ReadDarkNetWeights(os.path.join(os.getcwd(),'weights/darknet.weights'))
#reshape weights in every layer
for i in range(darkNet.layer_number):
l = darkNet.layers[i]
if(l.type == 'CONVOLUTIONAL'):
weight_array = l.weights
n = weight_array.shape[0]
weight_array = weight_array.reshape((n//(l.size*l.size),(l.size*l.size)))[:,::-1].reshape((n,))
weight_array = np.reshape(weight_array,[l.n,l.c,l.size,l.size])
l.weights = weight_array
if(l.type == 'CONNECTED'):
weight_array = l.weights
weight_array = np.reshape(weight_array,[l.input_size,l.output_size])
for i in range(98):
for j in range(20):
if(boxes[i].probs[j] != 0):
print i,j
print boxes[i].probs[j]
draw_detections(os.path.join(os.getcwd(),'images/dog.jpg'),98,0.2,boxes,20,labels,'dog.jpg')
'''
#for each image, we generate a detection result
imagePath = os.path.join(os.getcwd(), 'images')
images = [f for f in listdir(imagePath) if isfile(join(imagePath, f))]
for image_name in images:
timer = Timer()
image = crop(os.path.join(imagePath, image_name),
resize_width=512,
resize_height=512,
new_width=448,
new_height=448)
image = np.expand_dims(image, axis=0)
timer.tic()
out = model.predict(image)
timer.toc()
print('Total time is {:.3f}s ').format(timer.total_time)
predictions = out[0]
boxes = convert_yolo_detections(predictions)
boxes = do_nms_sort(boxes, 98)
draw_detections(os.path.join(imagePath, image_name), 98, 0.2, boxes, 20,
labels, image_name)
else:
print "Error: Unknown Layer Type", l.type
return model
def get_activations(model, layer, X_batch):
get_activations = theano.function(
[model.layers[0].input],
model.layers[layer].get_output(train=False),
allow_input_downcast=True)
activations = get_activations(X_batch) # same result as above
return activations
#image = readImg(os.path.join(os.getcwd(),'images/dog.file'))
image = crop(os.path.join(os.getcwd(), 'images/eagle.jpg'))
image = np.expand_dims(image, axis=0)
darkNet = ReadDarkNetWeights(
os.path.join(os.getcwd(), 'weights/darknet.weights'))
#reshape weights in every layer
for i in range(darkNet.layer_number):
l = darkNet.layers[i]
if (l.type == 'CONVOLUTIONAL'):
weight_array = l.weights
n = weight_array.shape[0]
weight_array = weight_array.reshape(
(n // (l.size * l.size), (l.size * l.size)))[:, ::-1].reshape(
(n, ))
weight_array = np.reshape(weight_array, [l.n, l.c, l.size, l.size])
l.weights = weight_array
print "Error: Unknown Layer Type"
model.add(Activation('softmax'))
return model
def get_activations(model, layer, X_batch):
get_activations = theano.function(
[model.layers[0].input],
model.layers[layer].get_output(train=False),
allow_input_downcast=True)
activations = get_activations(X_batch) # same result as above
return activations
#image = readImg(os.path.join(os.getcwd(),'images/dog.file'))
image = crop(os.path.join(os.getcwd(), 'images/dog.jpg'))
image = np.expand_dims(image, axis=0)
googleNet = ReadGoogleNetWeights(
os.path.join(os.getcwd(), 'weights/extraction.weights'))
#reshape weights in every layer
for i in range(googleNet.layer_number):
l = googleNet.layers[i]
if (l.type == 'CONVOLUTIONAL'):
weight_array = l.weights
n = weight_array.shape[0]
weight_array = weight_array.reshape(
(n // (l.size * l.size), (l.size * l.size)))[:, ::-1].reshape(
(n, ))
weight_array = np.reshape(weight_array, [l.n, l.c, l.size, l.size])
l.weights = weight_array
