What's the replacement for ScipyOptimizerInterface in Tensorflow 2?

Question

I want to experiment with Manumorph - A Tensorflow implementation of Deep Painterly Harmonization .

Unfortunately, I had zero success so far. I followed the instructions, but the current version of Tensorflow is not compatible with this implementation. I already updated the code with the Tensorflow tool tf_upgrade_v2.py . The script reported that tensorflow. contrib tensorflow. contrib is not supported anymore.

I also tried installing Tensorflow 1.15 but this version gives me DLL errors that I don't know how to solve.

I am new to Tensorflow and don't know what any of the code does. I hope someone with experience in Tensorflow can tell how I can make the line

train_step = tf.contrib.opt.ScipyOptimizerInterface(loss, 
                        var_list=[combination_im], 
                        options={'maxfun':20})

in main.py work with Tensorflow 2.

Answer 1

(Left this same answer in Use Scipy Optimizer with Tensorflow 2.0 for Neural Network training )

If you'd like something even more drag-and-drop for interfacing TF2 with solvers from scipy.optimize.minimize solver) you could take a look at this and see if it fits your use case:

• https://pypi.org/project/kormos/
• https://github.com/mbhynes/kormos