Something confused with the D_basic() function?

[hufengshuo07]

I tried read images directly as the input of D_basic() function. But no matter which kinds of images I read (real face, fake face ,or even boats), the outputs are always 0.00xxxx or 0.01xxxx. Has someone ever tried this? I can't figure it out.

Here's the session code:

with tf.Session(config=tf.ConfigProto(allow_soft_placement=True)) as sess:
img_placeholder1 = tf.placeholder(tf.float32, [1, 3, 1024, 1024])
img_placeholder2 = tf.placeholder(tf.float32, [1, 3, 1024, 1024])

img_now = cv2.imread('/media/sure/H/image_restoration/Compute-Features-master/test_images/Places365_test_00000002.jpg')/255.0*2-1
img_now = cv2.resize(img_now, (1024, 1024))

img_now = cv2.GaussianBlur(img_now, (9, 9), 2.0)
imgsmall = cv2.resize(img_now, (int(1024/ 16), int(1024/ 16)))
imgsmall = cv2.resize(imgsmall, (1024,1024))    #degraded
img_now = np.expand_dims(np.transpose(img_now,(2,0,1)), 0)
imgsmall = np.expand_dims(np.transpose(imgsmall,(2,0,1)), 0)

ddd1 = tf.nn.softplus(D_basic(img_placeholder1, None, is_training=False))
ddd2 = tf.nn.softplus(D_basic(img_placeholder2, None, is_training=False))

saver = tf.train.Saver()
saver.restore(sess,'./pretrain_model/gan_conv.ckpt')   #ckpt converted from official pkl

score1,score2 = sess.run([ddd1,ddd2], feed_dict={img_placeholder1:img_now,img_placeholder2:imgsmall})
print(score1,score2)

And got a result as ([0.01984515],[0.01300943]), which i think should be a value near 1.