Repository files navigation

TensorflowNoise2Noise implementation.

Noise2Noise is a machine learning algorithm that can learn signal reconstruction from onlynoisy examples, i.e. both inputs and targets are noisy realisations of the same image.

• Tfrecord files with jpeg encoded images under keyimage/encoded for training and evaluation.

(requiresnvidia-docker)

Build docker image

$ docker build -t n2n.

Run a command inside docker container

$ ./scripts/run-in-docker<command>

Mount extra volumes for input or output reasons. Current directory is already shared.

$ VOLUMES="/vol1:/vol1 /vol2:/vol2" ./scripts/run-in-docker<command>

$ python -m n2n.train --helpfull

Images from left to right are input image, denoised image, and ground truth noise free image.

$ python -m n2n.train<required-args> --noise additive_gaussian --loss l2

TODO: Compare on similar images.

$ python -m n2n.train<required-args> --noise additive_gaussian --loss l2 --adv_loss lsgan

$ python -m n2n.train<required-args> --noise text --loss l1

$ python -m n2n.train<required-args> --noise impulse --loss l0

TODO

TODO

• Raytracing/raycasting noise?