need_convert = False
run_inference = True
if os.path.isdir(os.path.join(os.getcwd(), "tflite")) is False:
os.makedirs(os.path.join(os.getcwd(), "tflite"))
model_path = "./tflite/converted_model.tflite"
base_path = os.getcwd()
base_inference_path = os.path.join(base_path, "dataset")
inference_image_path = os.path.join(base_inference_path,
"t10k-images-idx3-ubyte")
inference_label_path = os.path.join(base_inference_path,
"t10k-labels-idx1-ubyte")
inference_label_obj = inference_labels(inference_label_path)
inference_image_obj = inference_images(inference_image_path)
raw_Data = []
raw_Label = []
quan_steps = inference_image_obj.get_images_number()
for i in range(quan_steps):
raw_Data.append(inference_image_obj.read_images(1)[0])
raw_Label.append(inference_label_obj.read_labels(1)[0])
batchsize = 128
train = tf.convert_to_tensor(np.array(raw_Data, dtype='float32'))
train_label = tf.convert_to_tensor(np.array(raw_Label, dtype='float32'))
dataset = tf.data.Dataset.from_tensor_slices(
(train, train_label)).batch(batchsize, drop_remainder=True)
from tensorflow.python.platform import gfile
from tensorflow.python.client import timeline
tf.compat.v1.disable_eager_execution()
if __name__ == "__main__":
print("Begin inference!")
#base_path = "/home/mnist_dataset"
base_path = os.getcwd()
base_inference_path = os.path.join(base_path, "dataset")
inference_image_path = os.path.join(base_inference_path,
"t10k-images-idx3-ubyte")
inference_label_path = os.path.join(base_inference_path,
"t10k-labels-idx1-ubyte")
inference_labels = inference_labels(inference_label_path)
inference_images = inference_images(inference_image_path)
input_image_size = int(inference_images.get_row_number()) * int(
inference_images.get_column_number())
right_count = 0
batchsize = 128
total_time_ms = 0
global_step = 0
config = tf.compat.v1.ConfigProto(intra_op_parallelism_threads=28,
inter_op_parallelism_threads=1)
with tf.compat.v1.Session(config=config) as sess:
# saver = tf.train.import_meta_graph(os.path.join(base_path,"train_data/checkPoint/trainModel.meta"))
# saver.restore(sess, tf.train.latest_checkpoint(os.path.join(base_path,"train_data/checkPoint")))
with gfile.FastGFile(