def init_net(is_train, imdb_classes, args):
"""
initilize the network here.
"""
if args.net == 'alexnet':
fasterRCNN = alexnet(imdb_classes,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'vgg11':
fasterRCNN = vgg(imdb_classes,
11,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'vgg13':
fasterRCNN = vgg(imdb_classes,
13,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'vgg16':
# fasterRCNN = vgg16(imdb_classes, pretrained=is_train, class_agnostic=args.class_agnostic)
fasterRCNN = vgg(imdb_classes,
16,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'vgg19':
fasterRCNN = vgg(imdb_classes,
19,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'res101':
fasterRCNN = resnet(imdb_classes,
101,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'res50':
fasterRCNN = resnet(imdb_classes,
50,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'res152':
fasterRCNN = resnet(imdb_classes,
152,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'res18':
fasterRCNN = resnet(imdb_classes,
18,
pretrained=is_train,
class_agnostic=args.class_agnostic)
elif args.net == 'res34':
fasterRCNN = resnet(imdb_classes,
34,
pretrained=is_train,
class_agnostic=args.class_agnostic)
else:
raise Exception("network is not defined")
fasterRCNN.create_architecture()
return fasterRCNN
load_name = 'models/res101/saba_20171219_train/faster_rcnn_1_100_833.pth'
result_dir = 'result_res101'
elif args.net == 'res50':
fasterRCNN = resnet(pascal_classes, 50, pretrained=False, class_agnostic=args.class_agnostic)
elif args.net == 'res152':
fasterRCNN = resnet(pascal_classes, 152, pretrained=False, class_agnostic=args.class_agnostic)
elif args.net == 'res18':
fasterRCNN = Resnet18(pascal_classes, 18, pretrained=False, class_agnostic=args.class_agnostic)
load_name = 'models/res18/saba_20171219_train/faster_rcnn_1_100_833.pth'
result_dir = 'result_res18'
elif args.net == 'alexnet':
fasterRCNN = alexnet(pascal_classes, pretrained=False, class_agnostic=args.class_agnostic)
load_name = 'models/alexnet/saba_20171219_train/faster_rcnn_1_100_833.pth'
result_dir = 'result_alexnet'
elif args.net == 'inceptionv3':
fasterRCNN = Inceptionv3(pascal_classes, pretrained=False, class_agnostic=args.class_agnostic)
load_name = 'models/inceptionv3/saba_20171219_train/faster_rcnn_1_100_833.pth'
result_dir = 'result_inceptionv3'
cfg.TEST.MAX_SIZE = 600
elif args.net == 'dense121':
fasterRCNN = Dense121(pascal_classes, pretrained=False, class_agnostic=args.class_agnostic)
load_name = 'models/dense121/saba_20171219_train/faster_rcnn_1_100_833.pth'
result_dir = 'result_dense121'
else:
gt_boxes = gt_boxes.cuda()
# make variable
im_data = Variable(im_data)
im_info = Variable(im_info)
num_boxes = Variable(num_boxes)
gt_boxes = Variable(gt_boxes)
if args.cuda:
cfg.CUDA = True
# initilize the network here.
if args.s_net == 'alexnet':
student_net = alexnet(imdb.classes,
pretrained=True,
class_agnostic=args.class_agnostic)
else:
print("student network is not defined")
pdb.set_trace()
if args.t_net == 'vgg16':
teacher_net = vgg16(imdb.classes,
pretrained=False,
class_agnostic=args.class_agnostic,
teaching=True)
elif args.t_net == 'res101':
teacher_net = resnet(imdb.classes,
101,
pretrained=False,
class_agnostic=args.class_agnostic,