SPyRiT: an open source package for single-pixel imaging based on deep learning

We provide here the code to reproduce the results reported in

JFJP Abascal, T Baudier, R Phan, A Repetti, N Ducros, "SPyRiT: an open source package for single-pixel imaging based on deep learning," Preprint (2024).

Preprint view (main PDF + supplemental document): https://hal.science/hal-04662876

Preprint download (main PDF): https://hal.science/hal-04662876/document

Contact: [email protected], CREATIS Laboratory, University of Lyon, France.

Get the code from Github

There are two options:

1. Clone the entire spyrit-examples repository, which contains code for some other papers.

   git clone https://github.com/openspyrit/spyrit-examples.git

2. Or use the sparse-checkout command to get only the code corresponding to this paper.

   git clone -n --depth=1 --filter=tree:0 https://github.com/openspyrit/spyrit-examples
   cd spyrit-examples
   git sparse-checkout set 2024_spyrit
   git checkout

Installation

1. Create a conda environment

   conda create --name spyrit_2024
   conda activate spyrit_2024

2. Install pytorch using conda. E.g.,

   conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia

   Visit https://pytorch.org/get-started/locally/ if you need a different installation.

3. Install SPyRiT and a few more packages (until release of v3, checkout the spyrit_dev branch):

   pip install spyrit
   pip install girder-client
   pip install scikit-image

Download the models and data

Run the download_data.py script from the 2024_spyrit subfolder

cd spyrit-examples/2024_spyrit/ 
python download_data.py

The ImageNet (ILSVRC2012) test and validation sets can be downloaded from this url. They must saved in ./data/ILSVRC2012/test/all and ./data/ILSVRC2012/val/all.

If the images are already available on your computer, just create a symbolic link

• Linux:

  ln -s <imagenet folder> /data/ILSVRC2012/ 

• Windows Powershell (run as Administrator):

  New-Item -ItemType SymbolicLink -Name \data\ILSVRC2012\ -Target <imagenet folder>

The directory structure should be as follows:

|---spyrit-examples
|   |---2024_spyrit
|   |   |---data
|   |   |   |---ILSVRC2012
|   |   |   |   |---test
|   |   |   |   |   |---all
|   |   |   |   |   |   |---ILSVRC2012_test_00000001.JPEG
|   |   |   |   |   |   |---
|   |   |   |   |---val
|   |   |   |   |   |---all
|   |   |   |   |   |   |---ILSVRC2012_val_00000001.JPEG
|   |   |   |   |   |   |---
|   |   |   |---tomato_slice_2_zoomx2_spectraldata.npz
|   |   |   |---images
|   |   |   |---
|   |   |---model
|   |   |   |---
|   |   |---stat
|   |   |   |---
|   |   |---recon
|   |   |   |---
|   |   |---aux_functions.py
|   |   |---download_data.py
|   |   |---figure_2.py
|   |   |---figure_3.py
|   |   |---figure_4.py
|   |   |---supplemental_figure_S1.py
|   |   |---
|   |   |---table_1.py
|   |   |---train.py
|   |   |---utility_dpgd.py

How to reproduce the results of the paper?

1. To reproduce the sampling masks, acquisition matrices, measurements, and images in Figure 2, run figure_2.py.

2. To reproduce the reconstructions in Figures 3 and 4, run figure_3.py and figure_4.py, respectively. All images will be saved in \2024_spyrit\recon\

3. Run table_1.py to reproduce the metrics in Table 1.

4. Run supplemental_figure_Sxx.py for xx in {1, ..., 8} to reproduce all the figures in the supplemental document.