How do u produced the best.h5 model?

[Praneeth1441]

I am not able to replicate "best.h5" model provided in this repo by training the code present in this repo. The model that i have produced is showing whitish thing(noise) in the output Enhanced images.

The output from "best.h5" model looks natural.

Can u Tell what is the network to be followed to generate best.h5 model you have provided in this repo.

[l1teng]

The enhanced results of the official PyTorch implementation also meet the "whitish thing" artifacts (especially on the edge regions).

I did only check the results in the readme file of TensorFlow implementation. They look good, which performs well and shows no blur effects in the edge regions.

I did several experiments on the topic (how the blur effects were raised), and I think this blur effects phenomenon sources from the proposed method (not the implementation manner). The ablation experiment results also prove that assumption.

So, the same as @Praneeth1441, I am curious in what manner did @tuvovan try to achieve this.

Looking forward to your reply.

Regards.

--
Teng Li
[email protected]