Question

我需要在张量流中实现sRGB伽玛曲线。但是我无法计算张量流中的元素明智条件。

在sRGB曲线中，如果该值小于或等于0.0031308，则它是线性运算：x * 12.95 如果该值大于0.0031308，则为伽玛校正： 1.055 * x ^（1 / 2.4）-0.055

我尝试使用tf.cond（image> 0.0031308，function1，function2），但返回错误。如果有人可以帮助我，我将非常感激！！！

Answer 1

也许您需要tf.where()和tf.greater()。例如：

import tensorflow as tf
import numpy as np

image = np.random.random_sample(size=(2,3,3,1))/100
print(image)

image_tf = tf.placeholder(shape=(None,3,3,1),dtype=tf.float32)

new_image = tf.where(tf.greater(image_tf,0.0031308)
                     ,1.055*tf.pow(image_tf,1/2.4) - 0.055
                     ,image_tf*12.95)

with tf.Session() as sess:
    print(sess.run(new_image,feed_dict={image_tf:image}))
# image
[[[[0.0048715 ]
   [0.00472688]
   [0.00138391]]

  [[0.00848472]
   [0.00055767]
   [0.00835372]]

  [[0.00872362]
   [0.00390934]
   [0.00795842]]]


 [[[0.00756143]
   [0.00494474]
   [0.00201968]]

  [[0.00350234]
   [0.0056558 ]
   [0.00602147]]

  [[0.00692543]
   [0.0045199 ]
   [0.00012196]]]]
# new image
[[[[0.05975685]
   [0.05832487]
   [0.01792167]]

  [[0.08960549]
   [0.0072218 ]
   [0.08867098]]

  [[0.09128823]
   [0.04970375]
   [0.08579818]]]


 [[[0.08282798]
   [0.06047264]
   [0.02615486]]

  [[0.04501574]
   [0.06712133]
   [0.07035124]]

  [[0.07787357]
   [0.05623017]
   [0.00157937]]]]

张量流中的sRGB伽玛曲线工具

1 个答案: