Based on the below paper : https://arxiv.org/abs/1511.08861
I want to implement a mix loss of ms-ssim and l1 in tensorflow. Based on the nvidia implementation for Caffe at https://github.com/NVlabs/PL4NN/blob/master/src/loss.py
The loss is mathematically represented as : Loss= α*(MS-ssim)+(1-α)*(Gσ) *(l1_norm)
where α is a value emperically set and G stands for Gaussian filter with standard deviation 'σ'.
According to the caffe implementation it calculates forward and backward computation but as tensorflow handles the backpropagation how can this be implemented in tensorflow.
Is there any open source work on the mix loss fucntion for image processing?