predictByCEAL.py~

import os
import numpy as np
import tensorflow as tf

from alexnet import AlexNet
from dataprocess import ImageDataGenerator
from datetime import datetime
from tensorflow.contrib.data import Iterator
from config import *

os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = '3'

with tf.device('/cpu:0'):
    val_data = ImageDataGenerator(rest_file,
                                  mode='inference',
                                  batch_size=test_batch_size,
                                  num_classes=num_classes,
                                  shuffle=False)
    iterator = Iterator.from_structure(val_data.data.output_types,
                                       val_data.data.output_shapes)
    next_batch = iterator.get_next()

validation_init_op = iterator.make_initializer(val_data.data)
x = tf.placeholder(tf.float32, [test_batch_size, 227, 227, 3])
y = tf.placeholder(tf.float32, [test_batch_size, num_classes])
keep_prob = tf.placeholder(tf.float32)
model = AlexNet(x, keep_prob, num_classes, train_layers)
score = model.fc8_softmax

with tf.name_scope("accuracy"):
    index=tf.argmax(score, 1)
    correct_pred = tf.equal(tf.argmax(score, 1), tf.argmax(y, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

saver = tf.train.Saver()
val_batches_per_epoch = int(np.floor(val_data.data_size / test_batch_size))

low_conf_outlines = {}
high_conf_outlines = list()
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    ckpt = tf.train.get_checkpoint_state(checkpoint_path)
    if ckpt and ckpt.model_checkpoint_path:
	ckpt_name = os.path.basename(ckpt.model_checkpoint_path)
	saver.restore(sess, os.path.join(checkpoint_path, ckpt_name))
	print "load model successfully!!"
    print("{} Start CEAL...".format(datetime.now()))
    sess.run(validation_init_op)
    test_acc = 0.
    test_count = 0
    for _ in range(val_batches_per_epoch):
        filename_batch, img_batch, label_batch = sess.run(next_batch)
        pred_index, pred_score, acc = sess.run([index, score, accuracy], feed_dict={x: img_batch,
                                            y: label_batch,
                                            keep_prob: 1.})
        for i in xrange(test_batch_size):
            max_pred_conf = np.max( pred_score[i] )
            if max_pred_conf < 0.1:
                low_conf_outlines[max_pred_conf] = filename_batch[i] + ' ' + str( np.argmax( label_batch[i] ) ) + '\n'
            elif max_pred_conf > 0.9:
                high_conf_outlines.append( filename_batch[i] + ' ' + str(pred_index[i])+'\n')
                
                test_acc += acc
                test_count += 1
                
    test_acc /= test_count
    print("{} Pseudo-labeling Accuracy = {:.4f}".format(datetime.now(), test_acc))

    if os.path.exists(pred_file):
        os.remove(pred_file)
    f=open(pred_file,'a')
    for line in high_conf_outlines:
        f.write( line )
    print 'Pseudo-labeled Image Number' + str( len( high_conf_outlines ) )
    
    sorted( low_conf_outlines )
    count = 0 ;
    for key, val in low_conf_outlines.items():
        if count == selectALNUM:
            break
        count += 1
        f.write( val )
    f.close()
    
    print 'Finish CEAL!'