Xtensorflow

XTensorflow

A high level tensorflow lib which can convert defined tf model to caffe and other engine;

NetWork Definition:

from Xtensorflow import *    
input =  tf.placeholder(tf.float32, [batch_size, 256])
xnet = Xtensorflow(input, is_train=is_training, weight_decay=weight_decay, model_name=name)
fc1 = xnet.fc_layer(0, [batch_size, 256], tf.nn.relu)
fc2 = xnet.fc_layer(fc1, [batch_size, 128], tf.nn.relu)
fc3 = xnet.fc_layer(fc2, [batch_size, classnum], None)
output = xnet.get_network_output()

Convert to Caffemodel:

xnet.create_prototxt_and_caffemodel(sess,'model.prototxt','model.caffemodel')

Predefined Model Zoo：

input_imgs = tf.placeholder(tf.float32, [None, 224, 224, 3])
output, xnet = Resnet50(input_imgs, None, name='resnet50')
xnet_variables = tf.get_collection('xnet_varibale')
init = tf.variables_initializer(xnet_variables)
with tf.Session() as sess:
    sess.run(init)
    xnet.create_prototxt_and_caffemodel(sess, 'model.prototxt', 'model.caffemodel')