def reduce_model(model_path):
from kito import reduce_keras_model
from keras.models import load_model
m = load_model(model_path)
m_red = reduce_keras_model(m)
m_red.save(model_path[:-3] + '_reduced.h5')
def createImageEmbeddingModel():
import keras, kito
model = keras.applications.MobileNetV2()
model.summary()
outLayer = model.get_layer('global_average_pooling2d_1')
#return keras.Model(model.inputs, outLayer.output)
return kito.reduce_keras_model(keras.Model(model.inputs, outLayer.output))
def __init__(self,
img_size=64,
depth=16,
width=8,
pretrained_model="weights.29-3.76_utk.hdf5"):
self.img_size = img_size
self.model = WideResNet(img_size, depth=depth, k=width)()
self.model.load_weights(pretrained_model)
self.model = reduce_keras_model(self.model)
img_list = pickle.load(img_f)
with open(output_path[1], 'rb') as label_f:
label_list = pickle.load(label_f)
mean_arr = None # np.zeros(input_shape)
#for img in img_list:
# mean_arr += img.astype('float32')
#mean_arr /= len(img_list)
#print('mean shape:',mean_arr.shape, 'mean mean:',mean_arr.mean(), 'mean max:',mean_arr.max())
#mean_arr /= 255
#np.save('./mean.npy', mean_arr)
if config.pause:
nsml.paused(scope=locals())
bTrainmode = False
if config.mode == 'train':
bTrainmode = True
#nsml.load(checkpoint='86', session='Zonber/ir_ph1_v2/204') #Nasnet Large 222
nsml.load(checkpoint='0',
session='Zonber/ir_ph2/222') #InceptionResnetV2 222
print('convert start model')
intermediate_layer_model = Model(
inputs=model.input[0],
outputs=model.get_layer('triplet_loss_layer').input[0])
model_r = reduce_keras_model(intermediate_layer_model)
model_r.summary()
print('convert complete reduce model')
bind_model(model_r)
print('binde reduce model complete')
nsml.save(0) # this is display model name at lb
merge_avg = False
model_list = get_best_model_list(5, MODELS_PATH_KERAS)
model_list_reduced = []
for model_path in model_list:
if model_path == '':
model_list_reduced.append(None)
continue
store_path = model_path[:-3] + '_reduced.h5'
if not os.path.isfile(store_path):
model, _ = Model_3D_pretrained_densenet121(input_shape=SHAPE_SIZE,
dropout_val=0.2,
out_channels=1)
print('Load weights: {}'.format(model_path))
model.load_weights(model_path)
model = reduce_keras_model(model, verbose=True)
model.save(store_path)
model = None
gc.collect()
clear_session()
model_list_reduced.append(store_path)
_, preproc_input = Model_3D_pretrained_densenet121(input_shape=SHAPE_SIZE,
dropout_val=0.1,
out_channels=1)
score1, score2, score3, out_feat_file = valid_model(
model_list_reduced, preproc_input)
merged_model = merge_models(
SHAPE_SIZE,
model_list_reduced,
MODELS_PATH_KERAS +
#log_dir = 'logs/logits_only_000/'
#classes_path = 'class/voc_classes.txt'
#anchors_path = 'anchors/tiny_yolo_anchors.txt'
classes_path = 'class/elderly_classes.txt'
anchors_path = 'anchors/elderly_anchors.txt'
class_names = get_classes(classes_path)
num_classes = len(class_names)
anchors = get_anchors(anchors_path)
#input_shape = (416,416) # multiple of 32, hw
#ima#ge_input = Input(shape=(416,416, 3))
image_input = Input(shape=(None, None, 3))
#h, w = input_shape
num_anchors = len(anchors)
model = yolo_body(image_input, 3, num_classes)
model.load_weights(model_path)
model.summary()
print(len(model.layers))
model_reduced = reduce_keras_model(model)
#model_reduced.save('model_data/416bnfuse_small_mobilenets2_trained_model.h5')
#model_reduced.save('model_data/416bnfuse_tiny_yolo.h5')
#model_reduced.save('model_data/bnfuse_med_tiny_yolo.h5')
#model_reduced.save('model_data/bnfuse_a2_ds_small_mobilenets2_trained_model.h5')
model_reduced.save(
'model_data/bnfuse_eld_small_mobilenets2_trained_weights_final.h5')
model_reduced.summary()
print(len(model_reduced.layers))