def __init__(self, gpu_id, model_def, model_weights, labelmap_file, data_shape=None):
caffe.set_device(gpu_id)
caffe.set_mode_gpu()
# Load the net in the test phase for inference, and configure input preprocessing.
self.net = caffe.Net(model_def, # defines the structure of the model
model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
if data_shape is None:
data_shape = self.net.blobs['data'].data.shape
else:
if isinstance (data_shape, int):
data_shape = (data_shape, data_shape)
self.data_shape = data_shape
self.transformer = caffe.io.Transformer({'data': (1, 3, data_shape[0], data_shape[1])})
# self.transformer = caffe.io.Transformer({'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1))
self.transformer.set_mean('data', np.array([127.5, 127.5, 127.5])) # mean pixel
# the reference model operates on images in [0,255] range instead of [0,1]
#self.transformer.set_raw_scale('data', 255)
self.transformer.set_input_scale('data', 0.007843)
self.transformer.set_raw_scale('data', 255)
# the reference model has channels in BGR order instead of RGB
self.transformer.set_channel_swap('data', (2, 1, 0))
# load PASCAL VOC labels
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
def __init__(self, gpu_id, model_def, model_weights, image_resize,
labelmap_file):
caffe.set_device(gpu_id) #gpu编号
caffe.set_mode_gpu() #gpu模式
self.image_resize = image_resize
# 定义网络
self.net = caffe.Net(
model_def, # 网络模型
model_weights, # 网络权重
caffe.TEST) # 测试模式 不使用 droupout
# 输入图片处理变换器: 'data' is the name of the input blob == net.inputs[0]
self.transformer = caffe.io.Transformer(
{'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1)) # 通道*宽*高
self.transformer.set_mean('data',
np.array([104, 117,
123])) # voc数据集 各个通道均值 mean pixel
# the reference model operates on images in [0,255] range instead of [0,1]
self.transformer.set_raw_scale('data', 255) #[0,1]转换成[0,255]
# the reference model has channels in BGR order instead of RGB
self.transformer.set_channel_swap('data', (2, 1, 0)) # RGB --->BGR
# 图片标签文件 load PASCAL VOC labels
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
def load_model(self):
caffe.set_device(5)
caffe.set_mode_gpu()
# load labelmap
file = open(self.labelmap_file, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), labelmap)
# load model
net = caffe.Net(self.model_def, self.model_weights, caffe.TEST)
# image preprocessing
if '320' in self.model_def:
img_resize = 320
else:
img_resize = 512
net.blobs['data'].reshape(1, 3, img_resize, img_resize)
transformer = caffe.io.Transformer(
{'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2, 0, 1))
transformer.set_mean('data', np.array([104, 117, 123])) # mean pixel
transformer.set_raw_scale(
'data', 255
) # the reference model operates on images in [0,255] range instead of [0,1]
transformer.set_channel_swap('data', (
2, 1,
0)) # the reference model has channels in BGR order instead of RGB
# im_names = os.listdir('examples/images')
return net, transformer, labelmap
def imgfeature (img):
# 图片预处理设置
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape}) # 设定图片的shape格式(1,3,28,28)
transformer.set_transpose('data', (2, 0, 1)) # 改变维度的顺序，由原始图片(28,28,3)变为(3,28,28)
transformer.set_mean('data', np.array([63.7, 79.3, 136.1])) # 减去均值，前面训练模型时没有减均值，这儿就不用
# transformer.set_raw_scale('data', 255) # 缩放到【0，255】之间
# transformer.set_channel_swap('data', (2, 1, 0)) # 交换通道，将图片由RGB变为BGR
# im = caffe.io.load_image(img) # 加载图片
net.blobs['data'].data[...] = transformer.preprocess('data', img) # 执行上面设置的图片预处理操作，并将图片载入到blob中
# 执行测试
out = net.forward()
file = open(labels_file, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), labelmap)
# 得到网络的最终输出结果
loc = net.blobs['detection_out'].data[0][0]
# labels = np.loadtxt(labels_filename, str, delimiter='\t') # 读取类别名称文件
#
# prob = net.blobs['softmax'].data[0].flatten() # 取出最后一层（Softmax）属于某个类别的概率值，并打印
# print (prob)
# order = prob.argsort()[1] # 将概率值排序，取出最大值所在的序号
# print( 'the class is:', labels[order] ) # 将该序号转换成对应的类别名称，并打印
return loc,labelmap
def draw(image, predictions, labelmap_file):
if labelmap_file:
labelmap = caffe_pb2.LabelMap()
file = open(labelmap_file, 'r')
text_format.Merge(str(file.read()), labelmap)
file.close()
for prediction in predictions:
x, y, w, h = prediction[0]
score = prediction[1]
cls = prediction[2]
x1 = max(0, np.floor(x + 0.5).astype(int))
y1 = max(0, np.floor(y + 0.5).astype(int))
x2 = min(image.shape[1], np.floor(x + w + 0.5).astype(int))
y2 = min(image.shape[0], np.floor(y + h + 0.5).astype(int))
cv2.rectangle(image, (x1,y1), (x2,y2), (255, 0, 0), 2)
cv2.putText(image, '{0} {1:.2f}'.format(get_label_name(labelmap, cls), score),
(x1, y1-6),
cv2.FONT_HERSHEY_SIMPLEX,
0.6, (0,0,255), 1,
cv2.LINE_AA)
print('class: {0}, score: {1:.2f}'.format(cls, score))
print('box coordinate x, y, w, h: {0}'.format(prediction[0]))
return image
def get_class_num():
lm_f = open(labelmap_file, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(lm_f.read()), labelmap)
class_num = len(labelmap.item)
lm_f.close()
return class_num
def __init__(self):
caffe.set_device(0)
caffe.set_mode_gpu()
labelmap_file = open('data/coco/labelmap_coco_minsun.prototxt', 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(labelmap_file.read()), self.labelmap)
# * Load the net in the test phase for inference, and configure input preprocessing.
model_def = 'models/VGGNet/coco/SSD_300x300/deploy.prototxt'
model_weights = 'models/VGGNet/coco/SSD_300x300/VGG_coco_SSD_300x300_iter_240000.caffemodel'
self.net = caffe.Net(
model_def, # defines the structure of the model
model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
# input pre-processing: 'data' is the name of the input blob == net.inputs[0]
self.transformer = caffe.io.Transformer(
{'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1))
self.transformer.set_mean('data', np.array([104, 117,
123])) # mean pixel
self.transformer.set_raw_scale(
'data', 255
) # the reference model operates on images in [0,255] range instead of [0,1]
self.transformer.set_channel_swap('data', (
2, 1,
0)) # the reference model has channels in BGR order instead of RGB
# ### 2. SSD detection
# Load an image. Set net to batch size of 1
image_resize = 300
self.net.blobs['data'].reshape(1, 3, image_resize, image_resize)
def __init__(self, model_def, model_weights, image_resize, labelmap_file,
caffe_root, data_dir):
#caffe.set_device(gpu_id)
#caffe_root = '/home/ubuntu/caffe_ssd'
os.chdir(caffe_root) #切换到caffe_root目录
sys.path.insert(0, os.path.join(caffe_root, 'python'))
import caffe
from caffe.proto import caffe_pb2
caffe.set_mode_cpu()
self.image_resize = image_resize
# Load the net in the test phase for inference, and configure input preprocessing.
self.net = caffe.Net(
model_def, # defines the structure of the model
model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
self.transformer = caffe.io.Transformer(
{'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1))
self.transformer.set_mean('data', np.array([104, 117,
123])) # mean pixel
# the reference model operates on images in [0,255] range instead of [0,1]
self.transformer.set_raw_scale('data', 255)
# the reference model has channels in BGR order instead of RGB
self.transformer.set_channel_swap('data', (2, 1, 0))
# load PASCAL VOC labels
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
def load_model():
# load PASCAL VOC labels
labelmap_file = 'data/VOC0712/labelmap_voc.prototxt'
file = open(labelmap_file, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), labelmap)
model_def = 'models/VGGNet/VOC0712/SSD_300x300/deploy.prototxt'
model_weights = 'models/VGGNet/VOC0712/SSD_300x300/VGG_VOC0712_SSD_300x300_iter_120000.caffemodel'
net = caffe.Net(
model_def, # defines the structure of the model
model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2, 0, 1))
transformer.set_mean('data', np.array([104, 117, 123])) # mean pixel
transformer.set_raw_scale(
'data', 255
) # the reference model operates on images in [0,255] range instead of [0,1]
transformer.set_channel_swap(
'data',
(2, 1,
0)) # the reference model has channels in BGR order instead of RGB
# set net to batch size of 1
image_resize = 300
net.blobs['data'].reshape(1, 3, image_resize, image_resize)
return net, transformer, labelmap
def __init__(self, gpu_id, model_def, model_weights, image_resize,
labelmap_file):
caffe.set_device(gpu_id)
caffe.set_mode_gpu()
self.image_resize = image_resize
# Load the net in the test phase for inference, and configure input preprocessing.
self.net = caffe.Net(
model_def, # defines the structure of the model
model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
self.transformer = caffe.io.Transformer(
{'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1))
self.transformer.set_input_scale('data', 0.017)
self.transformer.set_mean('data', np.array([103.94, 116.78,
123.68])) # mean pixel
# the reference model operates on images in [0,255] range instead of [0,1]
self.transformer.set_raw_scale('data', 255)
# the reference model has channels in BGR order instead of RGB
self.transformer.set_channel_swap('data', (2, 1, 0))
# load labels
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
def initalNetModel(self):
caffe.set_mode_gpu()
caffe.set_device(self.gpuId)
self.net = caffe.Net(str(self.deployFileName),
str(self.modelFileName), caffe.TEST)
with open(str(self.labelFileName), 'r') as f:
self.label_list = caffe_pb2.LabelMap()
text_format.Merge(str(f.read()), self.label_list)
def load_labels(label_map_file_path):
""" Load labels from labelmap file
:param label_map_file_path: Labelmap file
"""
label_map = caffe_pb2.LabelMap()
label_map_serialized = open(label_map_file_path, 'rt').read()
text_format.Merge(str(label_map_serialized), label_map)
labels = [str(item.display_name) for item in label_map.item]
return labels
def labelmap(labelmap_file, label_info):
labelmap = caffe_pb2.LabelMap()
for i in range(len(label_info)):
labelmapitem = caffe_pb2.LabelMapItem()
labelmapitem.name = label_info[i]['name']
labelmapitem.label = label_info[i]['label']
labelmapitem.display_name = label_info[i]['display_name']
labelmap.item.add().MergeFrom(labelmapitem)
with open(labelmap_file, 'w') as f:
f.write(str(labelmap))
def labelmap(example, result):
labelmap = caffe_pb2.LabelMap()
# iterate through the sample file
for i in range(len(example)):
item = caffe_pb2.LabelMapItem()
item.name = example[i]['name']
item.label = example[i]['label']
labelmap.item.add().MergeFrom(item)
# writes the mapping result to the result file
with open(result, 'w') as file:
file.write(str(labelmap))
def __init__(self, model_def, model_weights, image_resize, labelmap_file):
self.image_resize = image_resize
if not os.path.exists(model_weights):
print("MobileNetSSD_deploy.caffemodel does not exist,")
print("use merge_bn.py to generate it.")
exit()
self.net = caffe.Net(model_def, model_weights, caffe.TEST)
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
self.colors = sns.color_palette("hls", len(self.labelmap.item))
def declare_network(model_def, model_weights, labelmap_file, args):
file = open(labelmap_file, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), labelmap)
net = caffe.Net(model_def, model_weights, caffe.TEST)
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2, 0, 1))
transformer.set_mean('data', np.array(args["img_mean"]))
transformer.set_input_scale('data', args["img_input_scale"])
return net, transformer, labelmap
def main():
global THREAD_RUNNING
args = parse_args()
# print('Called with args:')
# print(args)
# Initialize Caffe
if args.cpu_mode:
print('Running Caffe in CPU mode')
caffe.set_mode_cpu()
else:
print('Running Caffe in GPU mode')
caffe.set_device(0)
caffe.set_mode_gpu()
net = caffe.Net(DEFAULT_PROTOTXT, DEFAULT_MODEL, caffe.TEST)
# Build the class (index/name) dictionary from labelmap file
lm_handle = open(DEFAULT_LABELMAP, 'r')
lm_map = caffe_pb2.LabelMap()
text_format.Merge(str(lm_handle.read()), lm_map)
cls_dict = {x.label: x.display_name for x in lm_map.item}
# Open camera
if args.use_file:
cap = cv2.VideoCapture(args.filename)
# ignore image width/height settings here
elif args.use_usb:
cap = open_cam_usb(args.video_dev, args.image_width, args.image_height)
elif args.use_rtsp:
cap = open_cam_rtsp(args.rtsp_uri, args.image_width, args.image_height,
args.rtsp_latency)
else: # By default, use the Jetson onboard camera
cap = open_cam_onboard(args.image_width, args.image_height)
if not cap.isOpened():
sys.exit('Failed to open camera!')
# Start the sub-thread, which is responsible for grabbing images
THREAD_RUNNING = True
th = threading.Thread(target=grab_img, args=(cap, ))
th.start()
# Grab image and do object detection (until stopped by user)
open_window(args.image_width, args.image_height)
read_cam_and_detect(net, cls_dict, args.conf_th)
# Terminate the sub-thread
THREAD_RUNNING = False
th.join()
cap.release()
cv2.destroyAllWindows()
def initCaffeModule(self):
os.chdir(self.caffe_root)
sys.path.insert(0, 'python')
caffe.set_device(0)
caffe.set_mode_gpu()
logging.basicConfig(
level=logging.DEBUG,
format=
'%(asctime)s %(filename)s[line:%(lineno)d] %(levelname)s %(message)s',
datefmt='%a, %d %b %Y %H:%M:%S',
filename='allCaffe.log',
filemode='w')
# labelmap_file = '../caffemodule/supervisormodule/labelmap_voc.prototxt'
file = open(self.labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
# model_def = '../caffemodule/supervisormodule/deploy.prototxt'
# model_weights = '../caffemodule/supervisormodule/VGG_VOC0712_SSD_300x300_iter_60000.caffemodel'
self.net = caffe.Net(
self.model_def, # defines the structure of the model
self.model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
self.transformer = caffe.io.Transformer(
{'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1))
self.transformer.set_mean('data', np.array([104, 117,
123])) # mean pixel
self.transformer.set_raw_scale(
'data', 255
) # the reference model operates on images in [0,255] range instead of [0,1]
self.transformer.set_channel_swap('data', (
2, 1,
0)) # the reference model has channels in BGR order instead of RGB
# set net to batch size of 1
# image_resize = 300
self.net.blobs['data'].reshape(1, 3, int(self.image_resize),
int(self.image_resize))
self.functions = TaskCollections(self.client_conf)
credentials = pika.PlainCredentials('ye', '123456')
parameters = pika.ConnectionParameters('192.168.0.23', 5672, '/',
credentials)
#这里要设置重连机制
counectCounts = 0
while (counectCounts < 3):
try:
connection = pika.BlockingConnection(parameters)
counectCounts = 3
except Exception, e:
time.sleep(1)
counectCounts += 1
print "reconnecting..."
def __init__(self):
labelmap_file = '/home/davidbutra/Escritorio/caffe/data/VOC0712/labelmap_voc.prototxt'
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
#Net parameters necesary
model_def = '/home/davidbutra/Escritorio/caffe/models/VGGNet/VOC0712/SSD_300x300/deploy.prototxt'
model_weights = '/home/davidbutra/Escritorio/caffe/models/VGGNet/VOC0712/SSD_300x300/VGG_VOC0712_SSD_300x300_iter_120000.caffemodel'
self.net = caffe.Net(
model_def, # defines the structure of the model
model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
status = 0
ic = None
# Initializing the Ice run-time.
ic = Ice.initialize(sys.argv)
properties = ic.getProperties()
self.lock = threading.Lock()
try:
obj = ic.propertyToProxy("Numberclassifier.Camera.Proxy")
print obj
self.camera = jderobot.CameraPrx.checkedCast(obj)
if self.camera:
self.image = self.camera.getImageData("RGB8")
self.height = self.image.description.height
self.width = self.image.description.width
else:
print 'Interface camera not connected'
except:
traceback.print_exc()
exit()
status = 1
if ic:
# Clean up
try:
print("destroy")
print ic
ic.destroy()
print ic
except:
traceback.print_exc()
status = 1
print("except")
def _get_label(self):
if hasattr(self, 'labelmap') and self.labelmap:
return self.labelmap
fp = open(self.label_net_path, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(fp.read()), labelmap)
num_labels = len(labelmap.item)
label_map = {}
for idx in xrange(0, num_labels):
label_map[
labelmap.item[idx].label] = labelmap.item[idx].display_name
#label_map = {0: u'background', 1: u'hat', 2: u'glasses', 3: u'top', 4: u'shorts', 5: u'skirt', 6: u'trousers', 7: u'bag', 8: u'shoes', 9: u'dress', 10: u'outwear', 11: u'one_piece'}
return label_map
def labelMapSet(dataSets):
# load datasets labels
if dataSets == 'anngic':
labelmap_file = '/caffe/train_data/anngic/labelmap_anngic.prototxt'
elif dataSets == 'bdd100k':
labelmap_file = '/caffe/train_data/bdd100k/labelmap_bdd100k.prototxt'
elif dataSets == 'voc':
labelmap_file = '/caffe/caffe-ssd/data/VOC0712/labelmap_voc.prototxt'
file = open(labelmap_file, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), labelmap)
return labelmap
def load_label_prototxt(prototxt_file):
from google.protobuf import text_format
from caffe.proto import caffe_pb2 as cpb2
with open(prototxt_file) as f:
labelmap_aux = cpb2.LabelMap()
text_format.Merge(str(f.read()), labelmap_aux)
num_labels = len(labelmap_aux.item)
labelmap = {}
for item in labelmap_aux.item:
index = item.label
label = item.display_name
labelmap[index] = label
return labelmap
def __init__(self, gpu_id, model_def, model_weights, image_resize,
labelmap_file):
self.image_resize = image_resize
self.net = caffe.Net(model_def, model_weights, caffe.TEST)
self.transformer = caffe.io.Transformer(
{'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1))
self.transformer.set_mean('data', np.array([104, 117, 123]))
self.transformer.set_raw_scale('data', 255)
self.transformer.set_channel_swap('data', (2, 1, 0))
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
def __init__(self, gpu_id, model_def, model_weights, image_resize, labelmap_file):
#caffe.set_device(gpu_id)
#caffe.set_mode_gpu()
caffe.set_mode_cpu()
self.image_resize = image_resize
# Load the net in the test phase for inference, and configure input preprocessing.
self.net = caffe.Net(model_def, # defines the structure of the model
model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
# load PASCAL VOC labels
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
def get_classnames_from_labelmap(label_map_file):
"""Get classnames from labelmap"""
check_if_exist('Label map file', label_map_file)
labelmap = cpb2.LabelMap()
with open(label_map_file, 'r') as f:
text_format.Merge(str(f.read()), labelmap)
classnames = []
for item in labelmap.item:
classnames.append(str(item.display_name))
if len(classnames) == 0:
print('No classnames found in labelmap: {:s}.'.format(label_map_file))
sys.exit()
return classnames
def __init__(self, config, gpu_id=0, tracked=False):
# gpu preparation
if gpu_id >= 0:
caffe.set_device(gpu_id)
caffe.set_mode_gpu()
model_config = config['MODEL']
self.model_def = os.path.join(caffe_root, model_config['DEF'])
self.model_weights = os.path.join(caffe_root, model_config['WEIGHTS'])
self.labelmap = os.path.join(caffe_root, model_config['LABELMAP'])
self.is_tracked_results_ready = False
self.tracked = tracked
self.mot_tracker = None
self.mean_pixel = list([104, 117, 223])
self.num_classes = 32
self.colors = plt.cm.hsv(np.linspace(0, 1, self.num_classes)).tolist()
if self.tracked:
self.mot_tracker = Sort()
# load labelmap
with open(self.labelmap, 'r') as f:
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(f.read()), labelmap)
# load model
self.net = caffe.Net(self.model_def, self.model_weights, caffe.TEST)
# image preprocessing
if '320' in self.model_def:
img_resize = 320
else:
img_resize = 512
self.net.blobs['data'].reshape(1, 3, img_resize, img_resize)
self.transformer = caffe.io.Transformer(
{'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1))
self.transformer.set_mean('data',
np.array(self.mean_pixel)) # mean pixel
def init_model(self,caffe_root='/mnt/nndisk/tim/SmartVision/caffe-ssd'):
web_path = os.getcwd()
self.caffe_root = os.path.join(web_path, caffe_root)
# Make sure that caffe is on the python path:
os.chdir(self.caffe_root)
sys.path.insert(0, self.caffe_root + '/python') #####################
import caffe
self.caffe = caffe
if svs.compute_mode == "GPU":
self.caffe.set_device(0)
self.caffe.set_mode_gpu()
else:
self.caffe.set_mode_cpu()
####################################################
from caffe.proto import caffe_pb2
self.caffe_pb2 = caffe_pb2
# load PASCAL VOC labels
labelmap_file = 'data/coco/labelmap_coco.prototxt'
file = open(labelmap_file, 'r')
self.labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), self.labelmap)
self.capability = [C.OCCUPY_FOOTWAY_BY_CATERING]
####################################################
model_def = 'models/VGGNet/coco/SSD_512x512/deploy.prototxt' #####################
model_weights = 'models/VGGNet/coco/SSD_512x512/VGG_coco_SSD_512x512_iter_360000.caffemodel' #####################
self.net = caffe.Net(model_def, # defines the structure of the model
model_weights, # contains the trained weights
caffe.TEST) # use test mode (e.g., don't perform dropout)
# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
self.transformer = self.caffe.io.Transformer({'data': self.net.blobs['data'].data.shape})
self.transformer.set_transpose('data', (2, 0, 1))
self.transformer.set_mean('data', np.array([104,117,123])) # mean pixel
self.transformer.set_raw_scale('data', 255) # the reference model operates on images in [0,255] range instead of [0,1]
self.transformer.set_channel_swap('data', (2,1,0)) # the reference model has channels in BGR order instead of RGB
self.image_resize = 512
os.chdir(web_path)
print("model initialization done!")
def get_labelname(labelmap_file, labels):
#pdb.set_trace()
file = open(labelmap_file, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), labelmap)
num_labels = len(labelmap.item)
labelnames = []
if type(labels) is not list:
labels = [labels]
for label in labels:
found = False
for i in xrange(0, num_labels):
if label == labelmap.item[i].label:
found = True
labelnames.append(labelmap.item[i].display_name)
break
assert found == True
return labelnames
def show_MCDropout_result(transformed_image, resP6_mcdrop, resP5_mcdrop,
resP4_mcdrop, resP3_mcdrop):
# load MC-dropout model
model_def_bayes = os.path.abspath(
CUR_PATH + 'bayesian_dropout_model/deploy_multiple_input.prototxt')
model_weights_bayes = os.path.abspath(
CUR_PATH +
'../models/ResNet/coco/refinedet_resnet101_512x512/coco_refinedet_resnet101_512x512_iter_52000.caffemodel'
)
net_bayes = caffe.Net(model_def_bayes, model_weights_bayes, caffe.TEST)
# image preprocessing
if '320' in model_def_bayes:
img_resize = 320
else:
img_resize = 512
net_bayes.blobs['data'].reshape(1, 3, img_resize, img_resize)
net_bayes.blobs['data'].data[...] = transformed_image
net_bayes.blobs['dataP6'].data[...] = resP6_mcdrop
net_bayes.blobs['dataP5'].data[...] = resP5_mcdrop
net_bayes.blobs['dataP4'].data[...] = resP4_mcdrop
net_bayes.blobs['dataP3'].data[...] = resP3_mcdrop
res_bayes = net_bayes.forward()
# load labelmap
labelmap_file = 'data/coco/labelmap_coco.prototxt'
file = open(labelmap_file, 'r')
labelmap = caffe_pb2.LabelMap()
text_format.Merge(str(file.read()), labelmap)
detections = res_bayes['detection_out']
det_label = detections[0, 0, :, 1]
det_conf = detections[0, 0, :, 2]
det_xmin = detections[0, 0, :, 3] * image.shape[1]
det_ymin = detections[0, 0, :, 4] * image.shape[0]
det_xmax = detections[0, 0, :, 5] * image.shape[1]
det_ymax = detections[0, 0, :, 6] * image.shape[0]
result = np.column_stack(
[det_xmin, det_ymin, det_xmax, det_ymax, det_conf, det_label])
ShowResults(image, None, result, labelmap, save_fig=False, show_fig=True)
return
def init_models():
global CLS_LABEL_LIST
global THRESHOLD
global URLFLAG
global IMAGE_SIZE
global ONE_BATCH_SIZE
ONE_BATCH_SIZE = 16
URLFLAG = True if args.urlfileName else False
THRESHOLD = args.threshold
IMAGE_SIZE = 320
caffe.set_mode_gpu()
caffe.set_device(args.gpu_id)
deployName = os.path.join(args.modelBasePath, args.deployFileName)
modelName = os.path.join(args.modelBasePath, args.modelName)
labelName = os.path.join(args.modelBasePath, args.labelFileName)
net_cls = caffe.Net(deployName, modelName, caffe.TEST)
with open(labelName, 'r') as f:
CLS_LABEL_LIST = caffe_pb2.LabelMap()
text_format.Merge(str(f.read()), CLS_LABEL_LIST)
return net_cls
