I tried training a Autoencoder (Neural Network) network with colour image dataset.

when i try getting back the output by giving a input image, sometimes i get the output(greyscale pixel values) as i wanted. But with the same code and same input image, Manytimes i get an output value as 1 for all the pixels in the output image(fully white image). I understand there is a logical problem may be in the way i run the code, or because of saving the weighs in the memory or any other issues. I am new to this research area. Experts,Kindly help me. Thanks in advance.

[Walter Roberson]

Please show your attempt.

One possibility is that you have not trained enough samples.