lanl · bwohlberg · Feb 8, 2023 · Dec 17, 2022 · Dec 17, 2022 · Dec 17, 2022
diff --git a/data b/data
diff --git a/docs/docs_requirements.txt b/docs/docs_requirements.txt
@@ -1,10 +1,10 @@
 -r ../requirements.txt
-sphinx>=4.2.0
+sphinx>=4.2.0,<=5.3.0
 sphinxcontrib-napoleon
 sphinxcontrib-bibtex
 sphinx-autodoc-typehints
 faculty-sphinx-theme
-jinja2<3.1.0  # temporary fix for jina2/nbconvert bug
+jinja2<3.1.0  # temporary fix for jinja2/nbconvert bug
 traitlets!=5.2.2  # temporary fix for ipython/traitlets#741
 nbsphinx
 ipython_genutils

diff --git a/docs/source/conf.py b/docs/source/conf.py
@@ -170,7 +170,7 @@ def patched_parse(self):
 
 # General information about the project.
 project = "SCICO"
-copyright = "2020-2022, SCICO Developers"
+copyright = "2020-2023, SCICO Developers"
 
 # The version info for the project you're documenting, acts as replacement for
 # |version| and |release|, also used in various other places throughout the

diff --git a/docs/source/examples.rst b/docs/source/examples.rst
@@ -6,7 +6,7 @@ Usage Examples
 .. toctree::
    :maxdepth: 1
 
-.. include:: include/exampledepend.rst
+.. include:: include/examplenotes.rst
 
 
 Organized by Application
@@ -43,12 +43,14 @@ Deconvolution
 
    examples/deconv_circ_tv_admm
    examples/deconv_tv_admm
+   examples/deconv_tv_padmm
    examples/deconv_tv_admm_tune
    examples/deconv_microscopy_tv_admm
    examples/deconv_microscopy_allchn_tv_admm
    examples/deconv_ppp_bm3d_admm
    examples/deconv_ppp_bm3d_pgm
    examples/deconv_ppp_dncnn_admm
+   examples/deconv_ppp_dncnn_padmm
    examples/deconv_ppp_bm4d_admm
    examples/deconv_modl_train_foam1
    examples/deconv_odp_train_foam1
@@ -105,6 +107,7 @@ Plug and Play Priors
    examples/deconv_ppp_bm3d_admm
    examples/deconv_ppp_bm3d_pgm
    examples/deconv_ppp_dncnn_admm
+   examples/deconv_ppp_dncnn_padmm
    examples/deconv_ppp_bm4d_admm
    examples/demosaic_ppp_bm3d_admm
    examples/superres_ppp_dncnn_admm
@@ -123,6 +126,7 @@ Total Variation
    examples/deconv_circ_tv_admm
    examples/deconv_tv_admm
    examples/deconv_tv_admm_tune
+   examples/deconv_tv_padmm
    examples/deconv_microscopy_tv_admm
    examples/deconv_microscopy_allchn_tv_admm
    examples/denoise_l1tv_admm
@@ -216,8 +220,10 @@ Proximal ADMM
 .. toctree::
    :maxdepth: 1
 
+   examples/deconv_tv_padmm
    examples/denoise_tv_multi
    examples/denoise_cplx_tv_nlpadmm
+   examples/deconv_ppp_dncnn_padmm
 
 
 Non-linear Proximal ADMM

diff --git a/docs/source/include/exampledepend.rst b/docs/source/include/exampledepend.rst
diff --git a/docs/source/include/examplenotes.rst b/docs/source/include/examplenotes.rst
@@ -0,0 +1,32 @@
+.. _example_depend:
+
+Example Dependencies
+--------------------
+
+Some examples use additional dependencies, which are listed in `examples_requirements.txt <https://github.com/lanl/scico/blob/main/examples/examples_requirements.txt>`_.
+The additional requirements should be installed via pip, with the exception of ``astra-toolbox``,
+which should be installed via conda:
+
+   ::
+
+      conda install -c astra-toolbox astra-toolbox
+      pip install -r examples/examples_requirements.txt  # Installs other example requirements
+
+The dependencies can also be installed individually as required.
+
+Note that ``astra-toolbox`` should be installed on a host with one or more CUDA GPUs to ensure
+that the version with GPU support is installed.
+
+
+Run Time
+--------
+
+Most of these examples have been constructed with sufficiently small test problems to
+allow them to run to completion within 5 minutes or less on a reasonable workstation.
+Note, however, that it was not feasible to construct meaningful examples of the training
+of some of the deep learning algorithms that complete within a relatively short time;
+the examples "CT Training and Reconstructions with MoDL" and "CT Training and
+Reconstructions with ODP" in particular are much slower, and can require multiple hours
+to run on a workstation with multiple GPUs.
+
+|
diff --git a/docs/source/install.rst b/docs/source/install.rst
@@ -51,19 +51,23 @@ The instructions above install a CPU-only version of SCICO. To install a version
 
 1. Follow the CPU only instructions, above
 
-2. Identify which version of jaxlib was installed
+2. Install the version of jaxlib with GPU support, as described in the `JAX installation
+   instructions  <https://github.com/google/jax#installation>`_.
+   In the simplest case, the appropriate command is
 
    ::
 
-      pip list | grep jaxlib
+      pip install -upgrade "jax[cuda]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
 
-3. Install the same version of jaxlib, but with GPU support.
-   For help with this, see `JAX with GPU support <https://github.com/google/jax#installation>`_.
-   The command will be something like
 
+   but it may be necessary to explicitly specify the ``jaxlib`` version if the most recent release is not yet
+   supported by SCICO (as specified in the ``requirements.txt`` file), or if using a version of CUDA older
+   than 11.4, or CuDNN older than 8.2, in which case the command would be of the form
    ::
 
-      pip install --upgrade "jaxlib==0.3.0+cuda11.cudnn805" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
+      pip install -upgrade "jaxlib==0.4.2+cuda11.cudnn82" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
+
+   with appropriate substitution of ``jaxlib``, CUDA, and CuDNN version numbers.
 
 
 

diff --git a/examples/jnb.py b/examples/jnb.py
@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-# Copyright (C) 2022 by SCICO Developers
+# Copyright (C) 2022-2023 by SCICO Developers
 # All rights reserved. BSD 3-clause License.
 # This file is part of the SCICO package. Details of the copyright and
 # user license can be found in the 'LICENSE' file distributed with the
@@ -100,7 +100,7 @@ def execute_notebook(fname):
 
     with open(fname) as f:
         nb = nbformat.read(f, as_version=4)
-    ep = ExecutePreprocessor()
+    ep = ExecutePreprocessor(timeout=None)
     try:
         t0 = timer()
         out = ep.preprocess(nb)

diff --git a/examples/makenotebooks.py b/examples/makenotebooks.py
@@ -55,7 +55,7 @@ def notebook_path(sname):
 argparser.add_argument(
     "--all",
     action="store_true",
-    help="Process all notebooks, without  checking timestamps. "
+    help="Process all notebooks, without checking timestamps. "
     "Has no effect when files to process are explicitly specified.",
 )
 argparser.add_argument(

diff --git a/examples/scriptcheck.sh b/examples/scriptcheck.sh
@@ -39,21 +39,12 @@ if [ ! $# -eq 0 ] ; then
     exit 2
 fi
 
-# Check for presence of Xvfb tool which is used to avoid plots being displayed.
-if [ ! "$(which Xvfb 2>/dev/null)" ]; then
-    msg="Warning: required tool Xvfb not found: functionality will be degraded"
-    echo $msg >&2
-    pid=0
-else
-    Xvfb :20 -screen 0 800x600x16 > /dev/null 2>&1 &
-    pid=$!
-    export DISPLAY=:20.0
-fi
-
 # Set environment variables and paths. This script is assumed to be run
 # from its root directory.
 export PYTHONPATH=$SCRIPTPATH/..
 export PYTHONIOENCODING=utf-8
+export MPLBACKEND=agg
+export PYTHONWARNINGS=ignore:Matplotlib:UserWarning
 d='/tmp/scriptcheck_'$$
 mkdir -p $d
 retval=0
@@ -117,9 +108,4 @@ done
 # Remove temporary script directory.
 rmdir $d
 
-# Kill Xvfb process if it was started.
-if [ $pid != 0 ]; then
-  kill $pid
-fi
-
 exit $retval
diff --git a/examples/scripts/README.rst b/examples/scripts/README.rst
@@ -39,7 +39,9 @@ Deconvolution
    `deconv_circ_tv_admm.py <deconv_circ_tv_admm.py>`_
       Circulant Blur Image Deconvolution with TV Regularization
    `deconv_tv_admm.py <deconv_tv_admm.py>`_
-      Image Deconvolution with TV Regularization
+      Image Deconvolution with TV Regularization (ADMM Solver)
+   `deconv_tv_padmm.py <deconv_tv_padmm.py>`_
+      Image Deconvolution with TV Regularization (Proximal ADMM Solver)
    `deconv_tv_admm_tune.py <deconv_tv_admm_tune.py>`_
       Image Deconvolution Parameter Tuning
    `deconv_microscopy_tv_admm.py <deconv_microscopy_tv_admm.py>`_
@@ -51,7 +53,9 @@ Deconvolution
    `deconv_ppp_bm3d_pgm.py <deconv_ppp_bm3d_pgm.py>`_
       PPP (with BM3D) Image Deconvolution (APGM Solver)
    `deconv_ppp_dncnn_admm.py <deconv_ppp_dncnn_admm.py>`_
-      PPP (with DnCNN) Image Deconvolution
+      PPP (with DnCNN) Image Deconvolution (ADMM Solver)
+   `deconv_ppp_dncnn_padmm.py <deconv_ppp_dncnn_padmm.py>`_
+      PPP (with DnCNN) Image Deconvolution (Proximal ADMM Solver)
    `deconv_ppp_bm4d_admm.py <deconv_ppp_bm4d_admm.py>`_
       PPP (with BM4D) Volume Deconvolution
    `deconv_modl_train_foam1.py <deconv_modl_train_foam1.py>`_
@@ -121,7 +125,9 @@ Plug and Play Priors
    `deconv_ppp_bm3d_pgm.py <deconv_ppp_bm3d_pgm.py>`_
       PPP (with BM3D) Image Deconvolution (APGM Solver)
    `deconv_ppp_dncnn_admm.py <deconv_ppp_dncnn_admm.py>`_
-      PPP (with DnCNN) Image Deconvolution
+      PPP (with DnCNN) Image Deconvolution (ADMM Solver)
+   `deconv_ppp_dncnn_padmm.py <deconv_ppp_dncnn_padmm.py>`_
+      PPP (with DnCNN) Image Deconvolution (Proximal ADMM Solver)
    `deconv_ppp_bm4d_admm.py <deconv_ppp_bm4d_admm.py>`_
       PPP (with BM4D) Volume Deconvolution
    `demosaic_ppp_bm3d_admm.py <demosaic_ppp_bm3d_admm.py>`_
@@ -144,9 +150,11 @@ Total Variation
    `deconv_circ_tv_admm.py <deconv_circ_tv_admm.py>`_
       Circulant Blur Image Deconvolution with TV Regularization
    `deconv_tv_admm.py <deconv_tv_admm.py>`_
-      Image Deconvolution with TV Regularization
+      Image Deconvolution with TV Regularization (ADMM Solver)
    `deconv_tv_admm_tune.py <deconv_tv_admm_tune.py>`_
       Image Deconvolution Parameter Tuning
+   `deconv_tv_padmm.py <deconv_tv_padmm.py>`_
+      Image Deconvolution with TV Regularization (Proximal ADMM Solver)
    `deconv_microscopy_tv_admm.py <deconv_microscopy_tv_admm.py>`_
       Deconvolution Microscopy (Single Channel)
    `deconv_microscopy_allchn_tv_admm.py <deconv_microscopy_allchn_tv_admm.py>`_
@@ -228,7 +236,7 @@ ADMM
    `deconv_circ_tv_admm.py <deconv_circ_tv_admm.py>`_
       Circulant Blur Image Deconvolution with TV Regularization
    `deconv_tv_admm.py <deconv_tv_admm.py>`_
-      Image Deconvolution with TV Regularization
+      Image Deconvolution with TV Regularization (ADMM Solver)
    `deconv_tv_admm_tune.py <deconv_tv_admm_tune.py>`_
       Image Deconvolution Parameter Tuning
    `deconv_microscopy_tv_admm.py <deconv_microscopy_tv_admm.py>`_
@@ -238,7 +246,7 @@ ADMM
    `deconv_ppp_bm3d_admm.py <deconv_ppp_bm3d_admm.py>`_
       PPP (with BM3D) Image Deconvolution (ADMM Solver)
    `deconv_ppp_dncnn_admm.py <deconv_ppp_dncnn_admm.py>`_
-      PPP (with DnCNN) Image Deconvolution
+      PPP (with DnCNN) Image Deconvolution (ADMM Solver)
    `deconv_ppp_bm4d_admm.py <deconv_ppp_bm4d_admm.py>`_
       PPP (with BM4D) Volume Deconvolution
    `sparsecode_admm.py <sparsecode_admm.py>`_
@@ -266,6 +274,26 @@ Linearized ADMM
        Comparison of Optimization Algorithms for Total Variation Denoising
 
 
+Proximal ADMM
+^^^^^^^^^^^^^
+
+    `deconv_tv_padmm.py <deconv_tv_padmm.py>`_
+       Image Deconvolution with TV Regularization (Proximal ADMM Solver)
+    `denoise_tv_multi.py <denoise_tv_multi.py>`_
+       Comparison of Optimization Algorithms for Total Variation Denoising
+    `denoise_cplx_tv_nlpadmm.py <denoise_cplx_tv_nlpadmm.py>`_
+       Complex Total Variation Denoising with NLPADMM Solver
+    `deconv_ppp_dncnn_padmm.py <deconv_ppp_dncnn_padmm.py>`_
+       PPP (with DnCNN) Image Deconvolution (Proximal ADMM Solver)
+
+
+Non-linear Proximal ADMM
+^^^^^^^^^^^^^^^^^^^^^^^^
+
+    `denoise_cplx_tv_nlpadmm.py <denoise_cplx_tv_nlpadmm.py>`_
+       Complex Total Variation Denoising with NLPADMM Solver
+
+
 PDHG
 ^^^^
 
@@ -274,7 +302,7 @@ PDHG
     `denoise_tv_multi.py <denoise_tv_multi.py>`_
        Comparison of Optimization Algorithms for Total Variation Denoising
     `denoise_cplx_tv_pdhg.py <denoise_cplx_tv_pdhg.py>`_
-       Complex Total Variation Denoising
+       Complex Total Variation Denoising with PDHG Solver
 
 
 PGM

diff --git a/examples/scripts/ct_abel_tv_admm.py b/examples/scripts/ct_abel_tv_admm.py
@@ -50,8 +50,8 @@
 """
 Set up ADMM solver object.
 """
-λ = 1.9e1  # L1 norm regularization parameter
-ρ = 4.9e1  # ADMM penalty parameter
+λ = 2.9e1  # L1 norm regularization parameter
+ρ = 8.5e1  # ADMM penalty parameter
 maxiter = 100  # number of ADMM iterations
 cg_tol = 1e-4  # CG relative tolerance
 cg_maxiter = 25  # maximum CG iterations per ADMM iteration

diff --git a/examples/scripts/ct_abel_tv_admm_tune.py b/examples/scripts/ct_abel_tv_admm_tune.py
@@ -100,7 +100,7 @@ def eval_params(config, reporter):
 """
 Define parameter search space and resources per trial.
 """
-config = {"lambda": tune.loguniform(1e-2, 1e3), "rho": tune.loguniform(1e-1, 1e3)}
+config = {"lambda": tune.loguniform(1e0, 1e2), "rho": tune.loguniform(1e1, 1e3)}
 resources = {"gpu": 0, "cpu": 1}  # gpus per trial, cpus per trial
 
 
@@ -165,4 +165,37 @@ def eval_params(config, reporter):
 fig.show()
 
 
+"""
+Plot parameter values visited during parameter search and corresponding
+reconstruction PSNRs.The best point in the parameter space is indicated
+in red.
+"""
+𝜌 = [t.config["rho"] for t in analysis.trials]
+𝜆 = [t.config["lambda"] for t in analysis.trials]
+psnr = [t.metric_analysis["psnr"]["max"] for t in analysis.trials]
+minpsnr = min(max(psnr), 20.0)
+𝜌, 𝜆, psnr = zip(*filter(lambda x: x[2] >= minpsnr, zip(𝜌, 𝜆, psnr)))
+fig, ax = plot.subplots(figsize=(10, 8))
+sc = ax.scatter(𝜌, 𝜆, c=psnr, cmap=plot.cm.plasma_r)
+fig.colorbar(sc)
+plot.plot(
+    best_config["lambda"],
+    best_config["rho"],
+    ptyp="loglog",
+    lw=0,
+    ms=12.0,
+    marker="2",
+    mfc="red",
+    mec="red",
+    fig=fig,
+    ax=ax,
+)
+ax.set_xscale("log")
+ax.set_yscale("log")
+ax.set_xlabel(r"$\rho$")
+ax.set_ylabel(r"$\lambda$")
+ax.set_title("PSNR at each sample location\n(values below 20 dB omitted)")
+fig.show()
+
+
 input("\nWaiting for input to close figures and exit")