CODEWITHSUNDEEP
tensorflow
jkschin
jkschin

Reputation: 5844

How do we use tf.image.encode_jpeg to write out an image in TensorFlow?

Given a Tensor, I would like to use this function to get a 0D string and then write this out as "sample.jpg".

This can be easily achieved by OpenCV or Python PIL, but I would like to keep everything within TF if possible.

Upvotes: 3

Views: 8899

Answers (2)

Steven
Steven

Reputation: 5162

Alternatively, you can evaulate the image and use opencv or PIL to save it. 

import cv2 # opencv
from scipy.misc import imsave

... network generates an image ...
out = network(input)
img = tf.image.encode_jpeg(out)

img_to_save = img.eval(feed_dict)
# cv2.imwrite("myimage", img_to_save)
imsave("myimage.jpeg", img_to_save)

your image may or may not need a feed_dict to be evaluated.

Upvotes: 0

fabmilo
fabmilo

Reputation: 48330

In tensorflow the function would look like:

 import tensorflow as tf


def write_jpeg(data, filepath):
    g = tf.Graph()
    with g.as_default():
        data_t = tf.placeholder(tf.uint8)
        op = tf.image.encode_jpeg(data_t, format='rgb', quality=100)
        init = tf.initialize_all_variables()

    with tf.Session(graph=g) as sess:
        sess.run(init)
        data_np = sess.run(op, feed_dict={ data_t: data })

    with open(filepath, 'w') as fd:
        fd.write(data_np)


import numpy as np

R = np.zeros([128 * 128])
G = np.ones([128 * 128]) * 100
B = np.ones([128 * 128]) * 200

data = np.array(list(zip(R, G, B)), dtype=np.uint8).reshape(128, 128, 3)

assert data.shape == (128, 128, 3)

write_jpeg(data, "./test.jpeg")

the numpy part can be improved but it was for demonstration purposes only

Upvotes: 5

Related Questions