def inference(image, keep_prob):
print("setting up vgg initialized conv layers ...")
model_data = utils.get_model_data(FLAGS.model_dir, MODEL_URL)
mean = model_data['normalization'][0][0][0]
mean_pixel = np.mean(mean, axis=(0, 1))
weights = np.squeeze(model_data['layers'])
processed_image = utils.process_image(image, mean_pixel)
with tf.variable_scope("inference"):
image_net, pool_argmax = vgg_net(weights, processed_image)
conv_final_layer = image_net["pool5"]
fc_6 = utils.conv_layer(x=conv_final_layer,
W_shape=[7, 7, 512, 4096],
b_shape=4096,
name='fc_6',
padding='SAME')
fc_7 = utils.conv_layer(x=fc_6,
W_shape=[1, 1, 4096, 4096],
b_shape=4096,
name='fc_7',
padding='SAME')
with tf.variable_scope("Deconv"):
deconv_fc_6 = utils.deconv_layer(fc_7, [7, 7, 512, 4096], 512,
'fc6_deconv')
#unpool_5 = utils.unpool_layer2x2(deconv_fc_6, pool_argmax[-1], tf.shape(image_net["conv5_3"]))
unpool_5 = utils.unpool_layer2x2_batch(
deconv_fc_6, pool_argmax[-1]
) # Use unpool_layer2x2_batch if the input image is a batch
deconv_5_3 = utils.deconv_layer(unpool_5, [3, 3, 512, 512], 512,
'deconv_5_3')
deconv_5_2 = utils.deconv_layer(deconv_5_3, [3, 3, 512, 512], 512,
'deconv_5_2')
deconv_5_1 = utils.deconv_layer(deconv_5_2, [3, 3, 512, 512], 512,
'deconv_5_1')
#unpool_4 = utils.unpool_layer2x2(deconv_5_1, pool_argmax[-2], tf.shape(image_net["conv4_3"]))
unpool_4 = utils.unpool_layer2x2_batch(deconv_5_1, pool_argmax[-2])
deconv_4_3 = utils.deconv_layer(unpool_4, [3, 3, 512, 512], 512,
'deconv_4_3')
deconv_4_2 = utils.deconv_layer(deconv_4_3, [3, 3, 512, 512], 512,
'deconv_4_2')
deconv_4_1 = utils.deconv_layer(deconv_4_2, [3, 3, 256, 512], 256,
'deconv_4_1')
#unpool_3 = utils.unpool_layer2x2(deconv_4_1, pool_argmax[-3], tf.shape(image_net["conv3_3"]))
unpool_3 = utils.unpool_layer2x2_batch(deconv_4_1, pool_argmax[-3])
deconv_3_3 = utils.deconv_layer(unpool_3, [3, 3, 256, 256], 256,
'deconv_3_3')
deconv_3_2 = utils.deconv_layer(deconv_3_3, [3, 3, 256, 256], 256,
'deconv_3_2')
deconv_3_1 = utils.deconv_layer(deconv_3_2, [3, 3, 128, 256], 128,
'deconv_3_1')
#unpool_2 = utils.unpool_layer2x2(deconv_3_1, pool_argmax[-4], tf.shape(image_net["conv2_2"]))
unpool_2 = utils.unpool_layer2x2_batch(deconv_3_1, pool_argmax[-4])
deconv_2_2 = utils.deconv_layer(unpool_2, [3, 3, 128, 128], 128,
'deconv_2_2')
deconv_2_1 = utils.deconv_layer(deconv_2_2, [3, 3, 64, 128], 64,
'deconv_2_1')
#unpool_1 = utils.unpool_layer2x2(deconv_2_1, pool_argmax[-5], tf.shape(image_net["conv1_2"]))
unpool_1 = utils.unpool_layer2x2_batch(deconv_2_1, pool_argmax[-5])
deconv_1_2 = utils.deconv_layer(unpool_1, [3, 3, 64, 64], 64,
'deconv_1_2')
deconv_1_1 = utils.deconv_layer(deconv_1_2, [3, 3, 32, 64], 32,
'deconv_1_1')
score_1 = utils.deconv_layer(deconv_1_1, [1, 1, 21, 32], 21, 'score_1')
logits = tf.reshape(score_1, (-1, 21))
prediction = tf.argmax(tf.reshape(tf.nn.softmax(logits),
tf.shape(score_1)),
dimension=3)
print(prediction.shape, logits.shape, deconv_1_2.shape,
deconv_1_1.shape)
return prediction, logits
