def conver2resizeInput2pb():
# 从原来的h5文件中提取出只保存权重的文件
modelTrain = tf.keras.models.load_model('oldModel.h5')
modelTrain.save_weights('oldModelWeights.h5')
# 建立一个除了输入尺寸，结构层数和以前相同的网络用来加载权重
size =3648# cfg.TEST.INPUT_SIZE
# inputs = tf.keras.Input(shape=(3648, 5472, 3), name='in')
inputs = tf.keras.Input(shape=(size, size, 3), name='in')
imgDeconv, imgLab = mynet.Mynet(inputs,train=False)
modelW = tf.keras.Model(inputs, [imgDeconv, imgLab])
modelW.load_weights('oldModelWeights.h5')
# 建立一个 和 modelW共享权重，但是只输出lab的 用来训练的网络
network = tf.keras.Model(inputs, imgLab)
network.save("testmodel.h5")
img = np.zeros([1, size, size, 3], dtype=np.uint8)
img2 = img.astype(np.float32)
label = network.predict(img2)
# Convert Keras model to ConcreteFunction
full_model = tf.function(lambda x: network(x))
full_model = full_model.get_concrete_function(
tf.TensorSpec(network.inputs[0].shape, network.inputs[0].dtype))
# Get frozen ConcreteFunction
from tensorflow.python.framework.convert_to_constants import convert_variables_to_constants_v2
frozen_func = convert_variables_to_constants_v2(full_model)
frozen_func.graph.as_graph_def()
layers = [op.name for op in frozen_func.graph.get_operations()]
print("-" * 50)
print("Frozen model layers: ")
for layer in layers:
print(layer)
print("-" * 50)
print("Frozen model inputs: ")
print(frozen_func.inputs)
print("Frozen model outputs: ")
print(frozen_func.outputs)
# Save frozen graph from frozen ConcreteFunction to hard drive
tf.io.write_graph(graph_or_graph_def=frozen_func.graph,
logdir="./frozen_models",
name="frozen_graph.pb",
as_text=False)
def main(mode="test"):
if (sys.argv[1] == "train"):
module = mynet.Mynet().cuda()
train_dataset = datasets.MNIST(root='../data/',
train=True,
transform=transforms.ToTensor(),
download=True)
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=64,
shuffle=True)
mynet.train(100, train_loader, module)
if (sys.argv[1] == "test"):
module = torch.load(
"/home/zzl/zzlWorkspace/NNInference/MNIST/module/my_mnist.pkl")
test_dataset = datasets.MNIST(root='../data/',
train=False,
transform=transforms.ToTensor(),
download=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
batch_size=16,
shuffle=False)
mynet.test(test_loader, module)
def conver2resizeInput():
# 从原来的h5文件中提取出只保存权重的文件
modelTrain = tf.keras.models.load_model('oldModel.h5')
modelTrain.save_weights('oldModelWeights.h5')
# 建立一个除了输入尺寸，结构层数和以前相同的网络用来加载权重
size =3648# cfg.TEST.INPUT_SIZE
inputs = tf.keras.Input(shape=(3648, 5472, 3), name='in')
# inputs = tf.keras.Input(shape=(size, size, 3), name='in')
imgDeconv, imgLab = mynet.Mynet(inputs,train=False)
modelW = tf.keras.Model(inputs, [imgDeconv, imgLab])
modelW.load_weights('oldModelWeights.h5')
# 建立一个 和 modelW共享权重，但是只输出lab的 用来训练的网络
network = tf.keras.Model(inputs, imgLab)
network.save("testmodel.h5")
print(network.summary())
img = np.zeros([1, 3648, 5472, 3], dtype=np.uint8)
img2 = img.astype(np.float32)
label = network.predict(img2)
print(label.shape)
import os
import time
import shutil
import numpy as np
import tensorflow as tf
from config import cfg
import mynet
from VRDatasetC import NpDataset
size = cfg.TRAIN.INPUT_SIZE
inputs = tf.keras.Input(shape=(size, size, 1))
imgDeconv, imgLab = mynet.Mynet(inputs)
model = tf.keras.Model(inputs, [imgDeconv, imgLab])
keras.utils.plot_model(model, show_shapes=True, show_layer_names=True)
logdir = "./data/log"
optimizer = tf.keras.optimizers.Adam(0.0002)
if os.path.exists(logdir):
shutil.rmtree(logdir)
writer = tf.summary.create_file_writer(logdir)
# global_steps = tf.Variable(1, trainable=False, dtype=tf.int64)
# @tf.function
def learningModel(image_data, target):
with tf.GradientTape() as tape:
image_data = tf.cast(image_data, tf.float32)
target = tf.cast(target, tf.float32)
image_data = image_data / 255.0