sess.run(tf.global_variables_initializer())
saver.restore(sess, './model/S1_model_iter-0')
print("Starting training......")
for i in range(total_step):
imgs, labels = sess.run([imgs_batch, landmarks_batch])
sess.run(dan['S2_Optimizer'], feed_dict={x: imgs, y: labels})
if (i % auc_step == 0 or i == (total_step - 1)):
runTime = time.clock() - curTime
curTime = time.clock()
loss_s2 = dan['S2_Cost'].eval(feed_dict={x: imgs, y: labels})
#r = dan['r'].eval(feed_dict={x:imgs,y:labels})
print("step:%s loss_s2:%.4f Running time:%.2fs" %
(str(i).zfill(6), loss_s2, runTime))
if (i % auc_step == 0 or i == (total_step - 1)):
imgs_test, landmarks_test = loadTfrecords_t(tfrecordsTest, 1)
errors_s2 = []
for _ in range(100):
imgsTest, landmarksTest = sess.run([imgs_test, landmarks_test])
pred_s2 = sess.run(dan['S2_Ret'],
feed_dict={
x: imgsTest,
y: landmarksTest
})
error_s2 = LandmarkError(pred_s2[0], landmarksTest[0])
errors_s2.append(error_s2)
AUCError(errors_s2, threshold, step=0.0001, showCurve=False)
print(
sess.run(dan['S2_Cost'],
feed_dict={
x: imgsTest,
print("Starting training......")
for i in range(total_step):
imgs, labels = sess.run([imgs_batch, landmarks_batch])
sess.run(dan['s1_optimizer'], feed_dict={x: imgs, y: labels})
if (i % auc_step == 0 or i == (total_step - 1)):
runTime = time.clock() - curTime
curTime = time.clock()
loss_s1 = dan['s1_cost'].eval(feed_dict={
x: imgs,
y: labels
}) #,dan['S1_isTrain']:False,dan['S2_isTrain']:False})
print("step:%s loss_s1:%.6f Running time:%.2fs" %
(str(i).zfill(6), loss_s1, runTime))
if (i % auc_step == 0 or i == (total_step - 1)):
imgs_test, landmarks_test = loadTfrecords_t(
tfrecordsTest, batch_size)
errors_s1 = []
imgsTest, landmarksTest = sess.run([imgs_test, landmarks_test])
pred_s1 = sess.run(dan['s1_landmarks'],
feed_dict={
x: imgsTest,
y: landmarksTest
})
for i in range(batch_size):
error_s1 = LandmarkError(pred_s1[i],
landmarksTest[i],
normalization=['diagonal'])
errors_s1.append(error_s1)
AUCError(errors_s1, threshold, step=0.0001, showCurve=False)
print(
sess.run(dan['s1_cost'],