Denoise
usage: denoise [-h] [-d DEVICE] [-o OUTPUT] [--suffix SUFFIX]
[--format FORMAT_] [--normalize] [--stack]
[--save-prefix SAVE_PREFIX] [-m MODEL [MODEL ...]]
[-a DIR_A [DIR_A ...]] [-b DIR_B [DIR_B ...]] [--hdf HDF]
[--preload] [--holdout HOLDOUT] [--lowpass LOWPASS]
[--gaussian GAUSSIAN] [--inv-gaussian INV_GAUSSIAN]
[--deconvolve] [--deconv-patch DECONV_PATCH]
[--pixel-cutoff PIXEL_CUTOFF] [-s PATCH_SIZE]
[-p PATCH_PADDING] [--method {noise2noise,masked}]
[--arch {unet,unet-small,unet2,unet3,fcnet,fcnet2,affine}]
[--optim {adam,adagrad,sgd}] [--lr LR] [--criteria {L0,L1,L2}]
[-c CROP] [--batch-size BATCH_SIZE] [--num-epochs NUM_EPOCHS]
[--num-workers NUM_WORKERS] [-j NUM_THREADS]
[micrographs ...]
Positional Arguments
- micrographs
micrographs to denoise
Named Arguments
- -d, --device
which device to use, set to -1 to force CPU (default: 0)
Default: 0
- -o, --output
directory to save denoised micrographs
- --suffix
add this suffix to each output file name. if no output directory is specified, denoised micrographs are written to the same location as the input with a default suffix of “.denoised” (default: none)
Default: “”
- --format
output format for the images (default: mrc)
Default: “mrc”
- --normalize
normalize the micrographs
Default: False
- --stack
denoise a MRC stack rather than list of micorgraphs
Default: False
- --save-prefix
path prefix to save denoising model
- -m, --model
use pretrained denoising model(s). can accept arguments for multiple models the outputs of which will be averaged. pretrained model options are: unet, unet-small, fcnn, affine. to use older unet version specify unet-v0.2.1 (default: unet)
Default: [‘unet’]
- -a, --dir-a
directory of training images part A
- -b, --dir-b
directory of training images part B
- --hdf
path to HDF5 file containing training image stack as an alternative to dirA/dirB
- --preload
preload micrographs into RAM
Default: False
- --holdout
fraction of training micrograph pairs to holdout for validation (default: 0.1)
Default: 0.1
- --lowpass
lowpass filter micrographs by this amount (in pixels) before applying the denoising filter. uses a hard lowpass filter (i.e. sinc) (default: no lowpass filtering)
Default: 1
- --gaussian
Gaussian filter micrographs with this standard deviation (in pixels) before applying the denoising filter (default: 0)
Default: 0
- --inv-gaussian
Inverse Gaussian filter micrographs with this standard deviation (in pixels) before applying the denoising filter (default: 0)
Default: 0
- --deconvolve
apply optimal Gaussian deconvolution filter to each micrograph before denoising
Default: False
- --deconv-patch
apply spatial covariance correction to micrograph to this many patches (default: 1)
Default: 1
- --pixel-cutoff
set pixels >= this number of standard deviations away from the mean to the mean. only used when set > 0 (default: 0)
Default: 0
- -s, --patch-size
denoises micrographs in patches of this size. not used if < 1 (default: 1024)
Default: 1024
- -p, --patch-padding
padding around each patch to remove edge artifacts (default: 500)
Default: 500
- --method
Possible choices: noise2noise, masked
denoising training method (default: noise2noise)
Default: “noise2noise”
- --arch
Possible choices: unet, unet-small, unet2, unet3, fcnet, fcnet2, affine
denoising model architecture (default: unet)
Default: “unet”
- --optim
Possible choices: adam, adagrad, sgd
optimizer (default: adagrad)
Default: “adagrad”
- --lr
learning rate for the optimizer (default: 0.001)
Default: 0.001
- --criteria
Possible choices: L0, L1, L2
training criteria (default: L2)
Default: “L2”
- -c, --crop
training crop size (default: 800)
Default: 800
- --batch-size
training batch size (default: 4)
Default: 4
- --num-epochs
number of training epochs (default: 100)
Default: 100
- --num-workers
number of threads to use for loading data during training (default: 16)
Default: 16
- -j, --num-threads
number of threads for pytorch, 0 uses pytorch defaults, <0 uses all cores (default: 0)
Default: 0