nErrors = len(errors)
xAxis = list(np.arange(0., failureThreshold + step, step))
ced = [float(np.count_nonzero([errors <= x])) / nErrors for x in xAxis]
AUC = simps(ced, x=xAxis) / failureThreshold
failureRate = 1. - ced[-1]
print('AUC @ %.3f : %.3f' % (failureThreshold, AUC))
print('Failure rate: %.3f' % (failureRate))
print('Meanerror : %.3f' % (np.mean(errors)))
tfrecordsTrain = 'data/train.tfrecords'
tfrecordsTest = 'data/validation.tfrecords'
initmarks = np.load('data/initlandmark.npy')
initmarks = tf.constant(initmarks, dtype=tf.float32)
imgs_batch, landmarks_batch = loadTfrecords(tfrecordsTrain, batch_size)
#imgs_test, landmarks_test = loadTfrecords(tfrecordsTest ,batch_size)
x = tf.placeholder(dtype=tf.float32, shape=(None, 112, 112, 1))
y = tf.placeholder(dtype=tf.float32, shape=(None, 136))
training = FaceAlignmnetTraining(initmarks, batch_size, STAGE)
dan = training.createCNN(x, y, STAGE)
saver = tf.train.Saver()
#print(model.shape)
#tf.reset_default_graph()
with tf.Session() as sess:
curTime = time.clock()
saver = tf.train.Saver(
var_list=tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='S1'))
sess.run(tf.global_variables_initializer())
saver.restore(sess, './model/S1_model_iter-0')
print("Starting training......")
error = np.mean(np.sqrt(np.sum(
(gtLandmarks - resLandmarks)**2, axis=1))) / normDist
return error
def AUCError(errors, failureThreshold, step=0.0001, showCurve=False):
nErrors = len(errors)
xAxis = list(np.arange(0., failureThreshold + step, step))
ced = [float(np.count_nonzero([errors <= x])) / nErrors for x in xAxis]
AUC = simps(ced, x=xAxis) / failureThreshold
failureRate = 1. - ced[-1]
print('AUC @ %.3f : %.3f' % (failureThreshold, AUC))
print('Failure rate: %.3f' % (failureRate))
imgs_batch, landmarks_batch = loadTfrecords(tfrecordsPath, batch_size)
x = tf.placeholder(dtype=tf.float32, shape=(None, 112, 112, 1))
y = tf.placeholder(dtype=tf.float32, shape=(None, 136))
training = FaceAlignmnetTraining(initmarks, batch_size, STAGE)
dan = training.createCNN(x, y, l2_scale, STAGE)
saver = tf.train.Saver()
with tf.Session() as sess:
dataset = np.load(testSet)
dataImgs = (dataset["imgs"] - meanImg) / stdDevImg
#print(dataImgs.shape)
dataLandmarks = dataset["gtLandmarks"]
sess.run(tf.global_variables_initializer())
saver.restore(sess, outputsnap)
errors = []