def train():
train_file = ['data/train_cad.tfrecord']
valid_file = ['data/val_cad.tfrecord']
train_batch = batched_data(train_file, single_example_parser,
CONFIG.batch_size, 10 * CONFIG.batch_size)
valid_batch = batched_data(valid_file,
single_example_parser,
100,
shuffle=False)
with tf.Session() as sess:
myzfnet = ZFNet(CONFIG)
if CONFIG.mode == 'train0':
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
elif CONFIG.mode == 'train1':
myzfnet.restore(sess, CONFIG.model_save_path)
X_val = sess.run(valid_batch)
loss = []
acc = []
for epoch in range(1, CONFIG.epochs + 1):
X = sess.run(train_batch)
loss_, acc_, prediction_ = myzfnet.train(sess, X[0], X[1])
loss.append(loss_)
acc.append(acc_)
print(prediction_)
print(X[1])
print('>> %d/%d | loss: %f acc: %.2f%%' %
(epoch, CONFIG.epochs, loss_, 100.0 * acc_))
if epoch % CONFIG.per_save == 0:
acc_val = myzfnet.eval(sess, X_val[0], X_val[1])
print(' acc_val: %.2f%%\n' % (100.0 * acc_val))
myzfnet.save(sess, CONFIG.model_save_path)
def predict():
valid_file = ['data/val_cad.tfrecord']
valid_batch = batched_data(valid_file,
single_example_parser,
100,
shuffle=False)
with tf.Session() as sess:
myzfnet = ZFNet(CONFIG)
myzfnet.restore(sess, CONFIG.model_save_path)
X_val = sess.run(valid_batch)
result = myzfnet.predict(sess, X_val[0])
print(result)
print(
'----------------------------------------------------------------'
)