Odak

Odak (pronounced "O-dac") is the fundamental library for scientific computing in optical sciences, computer graphics, and visual perception. Odak is also the toolkit for the research conducted in Computational Light Laboratory. To learn more about what Odak can do to help your design, experimentation, and development, consult our documentation!

Getting Started

Installing

We encourage users to use virtual environments in their development pipeline when working with or developing odak. You can simply create and activate a virtual environment by using the following syntax:

python3 -m venv odak
source odak/bin/activate

Once activated, in the first usage, you can install odak using the previous instructions. To deactivate the virtual environemnt, you can always use deactivate command in your terminal. Once you have an activated virtual environment, please consider following any of the highlighted installation methods below.

For the most recent guidance on installing Odak, please consult installation documentation. Odak can be installed using pip:

pip3 install odak

or you can follow this, but remember that it will install the latest version in the repository this way:

pip3 install git+https://github.com/kaanaksit/odak --upgrade

or this:

git clone git@github.com:kaanaksit/odak.git
cd odak
pip3 install -r requirements.txt
pip3 install -e .

Usage and examples

You can import Odak and start designing your next in Optics, Computer Graphics, or Perception! We prepared a documentation on usage and much more. Absolute beginners can learn light, computation, and odak with our Computational Light course.

Here is a simple example of raytracing with odak:

import odak
import torch

starting_point = torch.tensor([0., 0., 0.])
end_point = torch.tensor([1., 1., 5.])
rays = odak.learn.raytracing.create_ray_from_two_points(
                                                        starting_point,
                                                        end_point
                                                       )

triangle = torch.tensor([[
                          [-5., -5., 5.],
                          [ 5., -5., 5.],
                          [ 0.,  5., 5.]
                         ]])

normals, distance, _, _, check = odak.learn.raytracing.intersect_w_triangle(
                                                                            rays,
                                                                            triangle
                                                                           )
print('intersection point is {}. Surface normal cosines are {}.'.format(normals[0, 0], normals[0, 1]))

Here is a simple example of computer-generated holography with odak:

import odak
import torch


wavelength = 532e-9
pixel_pitch = 8e-6 
distance = 5e-3
propagation_type = 'Angular Spectrum'
k = odak.learn.wave.wavenumber(wavelength)


amplitude = torch.zeros(500, 500)
amplitude[200:300, 200:300 ] = 1.
phase = torch.randn_like(amplitude) * 2 * odak.pi
hologram = odak.learn.wave.generate_complex_field(amplitude, phase)


image_plane = odak.learn.wave.propagate_beam(
                                             hologram,
                                             k,
                                             distance,
                                             pixel_pitch,
                                             wavelength,
                                             propagation_type,
                                             zero_padding = [True, False, True]
                                            )
image_intensity = odak.learn.wave.calculate_amplitude(image_plane) ** 2 
odak.learn.tools.save_image(
                            'image_intensity.png', 
                            image_intensity, 
                            cmin = 0., 
                            cmax = 1.
                           )

Here is a simple example of color conversion with odak:

import odak
import torch

input_rgb_image = torch.randn((1, 3, 256, 256))
ycrcb_image = odak.learn.perception.color_conversion.rgb_2_ycrcb(input_rgb_image)
rgb_image = odak.learn.perception.color_conversion.ycrcb_2_rgb(ycrcb_image)

Here is a simple example on deep learning methods with odak:

import odak
import torch


x1 = torch.arange(10).unsqueeze(-1) * 30.
pos_x1 = torch.arange(x1.shape[0]).unsqueeze(-1) * 1.
model_mlp = odak.learn.models.multi_layer_perceptron(
                                                     dimensions = [1, 5, 1],
                                                     bias = False,
                                                     model_type = 'conventional'
                                                    )


optimizer = torch.optim.AdamW(model_mlp.parameters(), lr = 1e-3)
loss_function = torch.nn.MSELoss()
for epoch in range(10000):
    optimizer.zero_grad()
    estimation = model_mlp(pos_x1)
    ground_truth = x1
    loss = loss_function(estimation, ground_truth)
    loss.backward(retain_graph = True)
    optimizer.step()
print('Training loss: {}'.format(loss.item()))


for item_id, item in enumerate(pos_x1):
    torch.no_grad()
    ground_truth = x1[item_id]
    estimation = model_mlp(item)
    print('Input: {}, Ground truth: {}, Estimation: {}'.format(item, ground_truth, estimation))

For more of these examples, you can either check our course documentation or visit our unit tests to get inspired.

Sample Projects that use Odak

Here are some sample projects that use odak:

• SpecTrack: Learned Multi-Rotation Tracking via Speckle Imaging
• Focal Surface Holographic Light Transport using Learned Spatially Adaptive Convolutions
• AutoColor: Learned Light Power Control for Multi-Color Holograms
• Multi-color Holograms Improve Brightness in Holographic Displays
• ChromaCorrect: Prescription Correction in Virtual Reality Headsets through Perceptual Guidance
• HoloBeam: Paper-Thin Near-Eye Displays
• Optimizing Vision and Visuals: Lectures on Cameras, Displays and Perception
• Realistic Defocus Blur for Multiplane Computer-Generated Holography
• Metameric Varifocal Computer-Generated Holography
• Learned Holographic Light Transport

How to cite

To add the link to this repository in your publication, please use Zenodo's citation. If you have used odak in your research project, please consider citing any of the following works:

@inproceedings{akcsit2023flexible,
  title={Flexible modeling of next-generation displays using a differentiable toolkit},
  author={Ak{\c{s}}it, Kaan and Kavakl{\i}, Koray},
  booktitle={Practical Holography XXXVII: Displays, Materials, and Applications},
  volume={12445},
  pages={131--132},
  year={2023},
  organization={SPIE}
}

@inproceedings{kavakli2022introduction,
  title={Introduction to Odak: a Differentiable Toolkit for Optical Sciences, Vision Sciences and Computer Graphics},
  author={Kavakl{\i}, Koray and Ak{\c{s}}it, Kaan},
  booktitle={Frontiers in Optics},
  pages={FTu1A--1},
  year={2022},
  organization={Optica Publishing Group}
}

@incollection{kavakli2022optimizing,
  title={Optimizing vision and visuals: lectures on cameras, displays and perception},
  author={Kavakli, Koray and Walton, David Robert and Antipa, Nick and Mantiuk, Rafa{\l} and Lanman, Douglas and Ak{\c{s}}it, Kaan},
  booktitle={ACM SIGGRAPH 2022 Courses},
  pages={1--66},
  year={2022}
}