Skip to content

Home

Jump to bottom
kumar saharsh edited this page Feb 23, 2020 · 1 revision

Welcome to the Neural-Style-Transfer wiki!

Problems I faced :

  1. The first thing was to write a custom loss function and minimize it. WE can't do this in Keras so options were to write in TensorFlow or fmin_l_bfgs_b in scipy. as I'm not very used coding in tensorflow i went with scipy.

  2. The generated image is very noisy and needs some other loss function to reduce the noise

  3. It's taking a lot of time and not scalable.

Clone this wiki locally