Deep-Alignment-Network-A-convolutional-neural-network-for-robust-face-alignment
Deep-Alignment-Network-A-convolutional-neural-network-for-robust-face-alignment copied to clipboard
zjjMaiMai
Freezing a model
Hi zjjMaiMai,
After I trained a model I have full check point file and graph.pbtxt, but when I want to freeze model, I cannot execute it.
I flow this:
bazel build tensorflow/tools:freeze_graph
bazel-bin/tensorflow/tools/freeze_graph
--input_graph=/tmp/model/graph.pbtxt
--input_checkpoint=/tmp/model/model.ckpt-1000
--output_graph=/tmp/frozen_graph.pb
--output_node_names="Sotfmax"
But it is error:
tensorflow.python.framework.errors_impl.NotFoundError: Op type not registered 'ImageProjectiveTransform' in binary running on diffdeep-server. Make sure the Op and Kernel are registered in the binary running in this process. Note that if you are loading a saved graph which used ops from tf.contrib, accessing (e.g.) tf.contrib.resampler should be done before importing the graph, as contrib ops are lazily registered when the module is first accessed
Please kindly help me, Thanks!
Hi @maidanglao99, now I do not work on this project. Hope you can fix it :)
import tensorflow as tf import os import shutil import json from DAN_V2 import vgg16_model_fn import numpy as np
""" The purpose of this script is to export the inference graph as a SavedModel.
Also it creates a .pb frozen inference graph. """
OUTPUT_FOLDER = 'export/' # for savedmodel PB_FILE_PATH = 'model.pb' GPU_TO_USE = '1' DATA_DIR = './trainset'
WIDTH, HEIGHT = 112,112 def read_dataset_info(data_dir): mean_shape = np.loadtxt(os.path.join(data_dir,'mean_shape.ptv'),dtype=np.float32,delimiter=',') imgs_mean = np.loadtxt(os.path.join(data_dir,'imgs_mean.ptv'),dtype=np.float32,delimiter=',') imgs_std = np.loadtxt(os.path.join(data_dir,'imgs_std.ptv'),dtype=np.float32,delimiter=',') return mean_shape.astype(np.float32) ,imgs_mean.astype(np.float32),imgs_std.astype(np.float32)
def export_savedmodel(): mean_shape,imgs_mean,imgs_std = read_dataset_info(DATA_DIR) config = tf.ConfigProto() #config.gpu_options.visible_device_list = GPU_TO_USE run_config = tf.estimator.RunConfig() run_config = run_config.replace( session_config=config ) estimator = tf.estimator.Estimator(vgg16_model_fn, model_dir='model_dir',config=run_config, params={ 'dan_stage':2, 'num_lmark':98, 'data_format': 'channels_last', 'batch_size': 1, 'multi_gpu': 0, 'mean_shape':mean_shape, 'imgs_mean':imgs_mean, 'imgs_std':imgs_std })
def serving_input_receiver_fn():
images = tf.placeholder(dtype=tf.uint8, shape=[None, HEIGHT, WIDTH, 1], name='images')
features = tf.to_float(images) * (1.0/255.0)
return tf.estimator.export.TensorServingInputReceiver(features=features, receiver_tensors={'images': images})
shutil.rmtree(OUTPUT_FOLDER, ignore_errors=True)
os.mkdir(OUTPUT_FOLDER)
estimator.export_savedmodel(OUTPUT_FOLDER, serving_input_receiver_fn)
def convert_to_pb():
subfolders = os.listdir(OUTPUT_FOLDER)
assert len(subfolders) == 1
last_saved_model = os.path.join(OUTPUT_FOLDER, subfolders[0])
graph = tf.Graph()
config = tf.ConfigProto()
#config.gpu_options.visible_device_list = GPU_TO_USE
with graph.as_default():
with tf.Session(graph=graph, config=config) as sess:
tf.saved_model.loader.load(sess, [tf.saved_model.tag_constants.SERVING], last_saved_model)
# output ops
keep_nodes = ['landmarks']
input_graph_def = tf.graph_util.convert_variables_to_constants(
sess, graph.as_graph_def(),
output_node_names=keep_nodes
)
output_graph_def = tf.graph_util.remove_training_nodes(
input_graph_def, protected_nodes=keep_nodes
)
#print(output_graph_def)
with tf.gfile.GFile(PB_FILE_PATH, 'wb') as f:
f.write(output_graph_def.SerializeToString())
print('%d ops in the final graph.' % len(output_graph_def.node))
tf.logging.set_verbosity('INFO') #export_savedmodel() #convert_to_pb()
print("start================================== \n") with tf.gfile.GFile("model.pb", 'rb') as f: graph_def = tf.GraphDef() graph_def.ParseFromString(f.read()) print([n.name for n in graph_def.node]) graph = tf.Graph() #print(graph)
config_proto = tf.ConfigProto() sess = tf.Session(graph=graph, config=config_proto)
with graph.as_default(): tf.import_graph_def(graph_def, name='import')
print("okokokokokokok") input_image = graph.get_tensor_by_name('import/images:0') output = graph.get_tensor_by_name('import/landmarks:0')
上面是我create pb的代码 请问您的代码是如何解决 AssertionError: landmarks is not in graph 这个问题的？ 我在使用了您的代码后出现了这个错误
I met the same problem,and bother me for a long time. do you solve this question?
