lensless 1.0.2

Package to control and image with a lensless camera running on a Raspberry Pi.

LenslessPiCam: A Hardware and Software Platform for Lensless Computational Imaging with a Raspberry Pi

• Setup.
• Data for examples.
• Reconstruction.
• Evaluating on a dataset.
• Remote capture.
• Remote display.
• Collecting MNIST.
• Citing this work.

This package provides functionalities to perform imaging with a lensless camera. We make use of a low-cost implementation of the DiffuserCam [1] where we use a piece of tape instead of the lens and the Raspberry Pi HQ camera sensor (V2 sensor is also supported). However, the same principles can be used for a different diffuser/mask and a different sensor (although the capture script would change). The idea of building a lensless camera from a Raspberry Pi and a piece of tape, and a dataset mentioned below come from Prof. Laura Waller's group at UC Berkeley:

• Build your own DiffuserCam tutorial.
• DiffuserCam lensless MIR Flickr dataset [2].

So a huge kudos to them for the idea and making the tools/code/data available!

We've also made a few Medium articles to guide users through the process of building the lensless camera, measuring data with it, and reconstruction. They are all laid out in this post.

Note that this material has been used for our graduate signal processing course at EPFL, and therefore includes some exercises / code to complete:

• lensless.autocorr.autocorr2d: to compute a 2D autocorrelation in the frequency domain,
• lensless.realfftconv.RealFFTConvolve2D: to pre-compute the PSF's Fourier transform, perform a convolution in the frequency domain with the real-valued FFT, and vectorize operations for RGB.

If you are an instructor, you can request access to the solutions here or send an email to eric[dot]bezzam[at]epfl[dot]ch.

Setup

The expected workflow is to have a local computer which interfaces remotely with a Raspberry Pi equipped with the HQ camera sensor (or V2 sensor as in the original tutorial).

The software from this repository has to be installed on your both your local machine and the Raspberry Pi (from the home directory of the Pi). Below are commands that worked for our configuration (Ubuntu 21.04), but there are certainly other ways to download a repository and install the library locally.

# download from GitHub
git clone [e-mail]:LCAV/LenslessPiCam.git

# install in virtual environment
cd LenslessPiCam
python3 -m venv lensless_env
source lensless_env/bin/activate
pip install -e .

# -- extra dependencies for local machine for plotting/reconstruction
pip install -r recon_requirements.txt

# (optional) try reconstruction on local machine
python scripts/recon/admm.py --psf_fp data/psf/tape_rgb.png \
--data_fp data/raw_data/thumbs_up_rgb.png --n_iter 5

On the Raspberry Pi, you may also have to install the following:

sudo apt-get install libimage-exiftool-perl
sudo apt-get install libatlas-base-dev

It is also possible to install the library via pip:

pip install lensless

Note that we highly recommend using Python 3.9, as its end-of-life is Oct 2025. Some Python library versions may not be available with earlier versions of Python.

For plotting on your local computer, you may also need to install Tk.

The scripts for remote capture and remote display assume that you can SSH to the Raspberry Pi without a password. To see this up you can follow instruction from this page. Do not set a password for your key pair, as this will not work with the provided scripts.

We have noticed problems with locale when running the remote capture and display scripts, for example:

perl: warning: Setting locale failed.
perl: warning: Please check that your locale settings:
...

This may arise due to incompatible locale settings between your local machine and the Raspberry Pi. There are two possible solutions to this, as proposed in this forum.

1. Comment SendEnv LANG LC_* in /etc/ssh/ssh_config on your laptop.
2. Comment AcceptEnv LANG LC_* in /etc/ssh/sshd_config on the Raspberry Pi.

Data for examples

You can download example PSFs and raw data that we've measured here. We recommend placing this content in the data folder.

You can download a subset for the DiffuserCam Lensless Mirflickr Dataset that we've prepared here with scripts/prepare_mirflickr_subset.py.

Reconstruction

The core algorithmic component of LenslessPiCam is the abstract class lensless.ReconstructionAlgorithm. The three reconstruction strategies available in LenslessPiCam derive from this class:

• lensless.GradientDescient: projected gradient descent with a non-negativity constraint. Two accelerated approaches are also available: lensless.NesterovGradientDescent and lensless.FISTA.
• lensless.ADMM: alternating direction method of multipliers (ADMM) with a non-negativity constraint and a total variation (TV) regularizer.
• lensless.APGD: accelerated proximal gradient descent with Pycsou as a backend. Any differentiable or proximal operator can be used as long as it is compatible with Pycsou, namely derives from one of DifferentiableFunctional or ProximableFunctional.

New reconstruction algorithms can be conveniently implemented by deriving from the abstract class and defining the following abstract methods:

• the update step: _update.
• a method to reset state variables: reset.
• an image formation method: _form_image.

One advantage of deriving from lensless.ReconstructionAlgorithm is that functionality for iterating, saving, and visualization is already implemented. Consequently, using a reconstruction algorithm that derives from it boils down to three steps:

1. Creating an instance of the reconstruction algorithm.
2. Setting the data.
3. Applying the algorithm.

For example, for ADMM:

    recon = ADMM(psf)
    recon.set_data(data)
    res = recon.apply(n_iter=n_iter)

A full running example can be run like so:

python scripts/recon/admm.py --psf_fp data/psf/tape_rgb.png \
--data_fp data/raw_data/thumbs_up_rgb.png --n_iter 5

A template for applying a reconstruction algorithm (including loading the data) can be found in scripts/recon/template.py.

Evaluating on a dataset

You can run ADMM on the DiffuserCam Lensless Mirflickr Dataset with the following script.

python scripts/evaluate_mirflickr_admm.py --data <FP>

where <FP> is the path to the dataset.

However, the original dataset is quite large (25000 files, 100 GB). So we've prepared this subset (200 files, 725 MB) which you can also pass to the script. It is also possible to set the number of files.

python scripts/evaluate_mirflickr_admm.py \
--data DiffuserCam_Mirflickr_200_3011302021_11h43_seed11 \
--n_files 10 --save

The --save flag will save a viewable image for each reconstruction.

You can also apply ADMM on a single image and visualize the iterative reconstruction.

python scripts/apply_admm_single_mirflickr.py \
--data DiffuserCam_Mirflickr_200_3011302021_11h43_seed11 \
--fid 172

Remote capture

You can remotely capture raw Bayer data with the following script.

python scripts/remote_capture.py --exp 0.1 --iso 100 --bayer --fp <FN> --hostname <HOSTNAME>

where <HOSTNAME> is the hostname or IP address of your Raspberry Pi, <FN> is the name of the file to save the Bayer data, and the other arguments can be used to adjust camera settings.

Remote display

For collecting images displayed on a screen, we have prepared some software to remotely display images on a Raspberry Pi installed with this software and connected to a monitor.

You first need to install the feh command line tool on your Raspberry Pi.

sudo apt-get install feh

Then make a folder where we will create and read prepared images.

mkdir ~/LenslessPiCam_display
cp ~/LenslessPiCam/data/original/mnist_3.png ~/LenslessPiCam_display/test.png

Then we can use feh to launch the image viewer.

feh LenslessPiCam_display --scale-down --auto-zoom -R 0.1 -x -F -Y

Then from your laptop you can use the following script to display an image on the Raspberry Pi:

python scripts/remote_display.py --fp <FP> --hostname <HOSTNAME> \
--pad 80 --vshift 10 --brightness 90

where <HOSTNAME> is the hostname or IP address of your Raspberry Pi, <FN> is the path on your local computer of the image you would like to display, and the other arguments can be used to adjust the positioning of the image and its brightness.

Collecting MNIST

We provide a couple scripts to collect MNIST with the proposed camera.

Script that can be launched from the Raspberry Pi:

python scripts/collect_mnist_on_device.py --input_dir MNIST_original \
--output_dir MNIST_meas

If the MNIST dataset is not available at MNIST_original it will be downloaded from here. The above command will measure the training set. The --test flag can be used to measure the test set. It is recommended to run the script from a screen session as it takes a long time to go through all the files! The --n_files <N_FILES> option can be used to measure a user-specified amount of files.

To remotely collect the MNIST dataset (although quite slow due to copying files back and forth):

python scripts/collect_mnist.py --hostname <IP_ADDRESS> --output_dir MNIST_meas

Citing this work

If you use these tools in your own research, please cite the following:

@misc{lenslesspicam,
	url = {https://infoscience.epfl.ch/record/294041?&ln=en},
	author = {Bezzam, Eric and Kashani, Sepand and Vetterli, Martin and Simeoni, Matthieu},
	title = {Lensless{P}i{C}am: A Hardware and Software Platform for Lensless Computational Imaging with a {R}aspberry {P}i},
	publisher = {Infoscience},
	year = {2022},
}

References

[1] Antipa, N., Kuo, G., Heckel, R., Mildenhall, B., Bostan, E., Ng, R., & Waller, L. (2018). DiffuserCam: lensless single-exposure 3D imaging. Optica, 5(1), 1-9.

[2] Monakhova, K., Yurtsever, J., Kuo, G., Antipa, N., Yanny, K., & Waller, L. (2019). Learned reconstructions for practical mask-based lensless imaging. Optics express, 27(20), 28075-28090.

[3] Boyd, S., Parikh, N., & Chu, E. (2011). Distributed optimization and statistical learning via the alternating direction method of multipliers. Now Publishers Inc.