DPReview.com is closing April 10th - Find out more

Demosaicnet adaptation for Xtrans raw conversion

Started Jul 12, 2021 | Discussions thread
ForumParentFirstPrevious
Flat view
robcuda New Member • Posts: 1
Demosaicnet adaptation for Xtrans raw conversion

With an upcoming new family member, I've become interested again in my Fuji X-E1. However, it still looks like generally speaking, using the in-camera JPEG is often the best option.

I did some digging and found an open-source project, demosaicnet (from 2016!), which is a machine-learning project for, well, demosaicing. It required some initial workflow with dcraw, but the results were particularly good in my opinion for Fuji XTrans raw conversion.

I've forked the project and modified it to accept RAF raw images and folders. With some additional code and insight (credited) I've also added some basic processing for black level, white level, and sRGB colorspace.

If you are comfortable using Python you can clone the repo at the link below. Note, running the pretrained machine-learning model on a CPU is very slow. If you have a GPU and can install CUDA drivers (and PyTorch with CUDA), it will be pretty quick.

https://github.com/robcuda/xtrans-develop

Sample:

Example X-E2 raw conversion; RawTherapee 5.8 screenshot on DSCF4199.RAF (dpreview-sourced)

Fujifilm X-E1 Fujifilm X-E2
If you believe there are incorrect tags, please send us this post using our feedback form.
ForumParentFirstPrevious
Flat view
ForumParentFirstPrevious
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow