# print("Loading vgg19 weights...")
# vgg_model = VGG19(include_top=False, weights='imagenet')
# for layer in model.layers:
# if layer.name in from_vgg:
# vgg_layer_name = from_vgg[layer.name]
# layer.set_weights(vgg_model.get_layer(vgg_layer_name).get_weights())
# print("Loaded VGG19 layer: " + vgg_layer_name)
last_epoch = 0
# prepare generators
if use_client_gen:
train_client = DataGeneratorClient(port=pargs.port,
host="localhost",
hwm=160,
batch_size=batch_size,
pstages=paramNumStages)
train_client.start()
train_di = train_client.gen()
train_samples = 52597
val_client = DataGeneratorClient(port=pargs.port + 1,
host="localhost",
hwm=160,
batch_size=batch_size,
pstages=paramNumStages)
val_client.start()
val_di = val_client.gen()
val_samples = 2645
else:
lc = 0
for layer in model.layers:
try:
rn_layer = rn.get_layer(layer.name)
if type(rn_layer) is Conv2D:
print "Loading weights for layer", layer.name
layer.set_weights(rn_layer.get_weights())
lc += 1
except:
print "Skipping Layer ", layer.name
print "Done loading weights for %d resnet conv layers" % lc
# prepare generators
stages = 1
train_client = DataGeneratorClient(port=5555, host="localhost", hwm=160, batch_size=batch_size, with_pafs=True, stages=stages)
train_client.start()
train_di = train_client.gen()
train_samples = 3000 # 52597 # All train samples in the COCO dataset
val_client = DataGeneratorClient(port=5556, host="localhost", hwm=160, batch_size=batch_size, with_pafs=True, stages=stages)
val_client.start()
val_di = val_client.gen()
val_samples = 1000 # 2645 # All validation samples in the COCO dataset
# euclidean loss as implemented in caffe https://github.com/BVLC/caffe/blob/master/src/caffe/layers/euclidean_loss_layer.cpp
def eucl_loss(x, y):
return K.sum(K.square(x - y)) / batch_size / 2
losses = {}
vgg_model = VGG19(include_top=False, weights='imagenet')
for layer in model.layers:
if layer.name in from_vgg:
vgg_layer_name = from_vgg[layer.name]
layer.set_weights(
vgg_model.get_layer(vgg_layer_name).get_weights())
print("Loaded VGG19 layer: " + vgg_layer_name)
last_epoch = 0
# prepare generators
if use_client_gen:
train_client = DataGeneratorClient(port=5555,
host="localhost",
hwm=160,
batch_size=10)
train_client.start()
train_di = train_client.gen()
train_samples = 52597
val_client = DataGeneratorClient(port=5556,
host="localhost",
hwm=160,
batch_size=10)
val_client.start()
val_di = val_client.gen()
val_samples = 2645
else:
train_di = DataIterator("../dataset/train_dataset.h5",
data_shape=(3, 368, 368),
vgg_model = VGG19(include_top=False, weights='imagenet')
for layer in model.layers:
if layer.name in from_vgg:
vgg_layer_name = from_vgg[layer.name]
layer.set_weights(
vgg_model.get_layer(vgg_layer_name).get_weights())
print("Loaded VGG19 layer: " + vgg_layer_name)
last_epoch = 0
# prepare generators
if use_client_gen:
train_client = DataGeneratorClient(port=5555,
host="localhost",
nb_parts=14,
nb_limbs=13,
hwm=160,
batch_size=batch_size)
train_client.start()
train_di = train_client.gen()
train_samples = 4000 # modified by g.hy
val_client = DataGeneratorClient(port=5556,
host="localhost",
nb_parts=14,
nb_limbs=13,
hwm=160,
batch_size=batch_size)
val_client.start()
val_di = val_client.gen()
val_samples = 500
print("Loading vgg19 weights...")
vgg_model = VGG19(include_top=False, weights='imagenet')
for layer in model.layers:
if layer.name in from_vgg:
vgg_layer_name = from_vgg[layer.name]
layer.set_weights(vgg_model.get_layer(vgg_layer_name).get_weights())
print("Loaded VGG19 layer: " + vgg_layer_name)
last_epoch = 0
# prepare generators
if use_client_gen:
train_client = DataGeneratorClient(port=5555, host="localhost", hwm=160, batch_size=10)
train_client.start()
train_di = train_client.gen()
train_samples = 52597
val_client = DataGeneratorClient(port=5556, host="localhost", hwm=160, batch_size=10)
val_client.start()
val_di = val_client.gen()
val_samples = 2645
else:
train_di = DataIterator("../dataset/train_dataset.h5", data_shape=(3, 368, 368),
mask_shape=(1, 46, 46),
label_shape=(57, 46, 46),
vec_num=38, heat_num=19, batch_size=batch_size, shuffle=True)
train_samples=train_di.N
val_di = DataIterator("../dataset/val_dataset.h5", data_shape=(3, 368, 368),
from keras.callbacks import LearningRateScheduler, ModelCheckpoint, CSVLogger, TensorBoard
from keras.layers.convolutional import Conv2D
from keras.applications.resnet50 import ResNet50
import keras.backend as K
from hm_model import acc_norm
batch_size = 100
train_samples = 2000
WEIGHTS_BEST = "vnect_weights.h5"
val_client = DataGeneratorClient(port=5556,
host="localhost",
hwm=160,
batch_size=batch_size,
with_pafs=False,
stages=1)
val_client.start()
val_di = val_client.gen()
val_samples = 2000
import vnect_model as md
model = md.get_training_model()
# load previous weights or vgg19 if this is the first run
if os.path.exists(WEIGHTS_BEST):
print("Loading the best weights...")
model.load_weights(WEIGHTS_BEST)
else:
raise ("Weights file %s not found " % WEIGHTS_BEST)
