kwasniok/python-nerte

non-euclidean ray tracing engine

python-nerte's Issues

make values immutable

Make all members of the values package immutable.

May not increase performance and blow up code.

feature/immutable-values

add tools to judge the convergence of an image

e.g. generate sequence of images with increasing resolution, max ray steps etc.

no automation yet, but filters and meta data help

remove tangent delta

replace by tuple of vector for velocity and vector for acceleration

allow to add vectors to coordinates

remove swirl segmented ray geometry

CON:

• this method does not converge
• might not be a proper geometry (metric)

PRO:

• results look nice
• fast
• good for demos

redesign miss reasons to reflect hierachy

an enum is not powerfull enough to provide the representation of the hierachy of miss reasons

hierachy is:

• uninitialized
• no intersection
• ray outside manifold
  +---+ray initialized outside of manifold (usefull to analyze embedding of canvas manifold)
  +---+ray reached outside of manifold (usefull to analyze propagation in manifold)

EDIT: Not completely true. Not worth the hasstle.

refine manifold doc

• cylindrical

• cylindrical swirl

• doc is not up to date

• explicit note for ValueError (coordinate restrictions)

make render progress responsive

• progress bar?
• update image on a per pixel patch basis (requires #40)

refactor image renderers

• move failure management to image renderer (failure color etc.)
• astract pixel rendering via hook mechanism

add ray info for intersections with objects

• include more information why an intersection failed
• add render modes to display that

branch: feature/ray_depth

allow to render subsection(submanifold) of canvas only

This speeds up test renders significantly.
It is also sometimes necessary to limit the domain to a non-rectangular subdomain.

image render does not test for ray depth

make metric property of matrix

a special type is redundant
check and change invertibility instead

fix initial ray generation for curved geometries

The initial tangent is currently approximated with the coordinate delta between start and target this is wrong in general!

• Runge-Kutta geometries
• segmented ray geometries

reduce intersection info miss reasons to one (optional) item

using a set of miss reasons is too cumbersome and does not serve any purpose

extend intersection info

• include optional parameters: steps

refactor ray segment into tangential vector

• ray segment has tangential vector
• ray segment delta
• remove ray segment

segment != tangent:
When using Runge-Kutta the next segments position might not differ from the previous by the direction of the previous. This is because the actual segment is curved while the discrete delta is straight. The direction of the previous direction is just the tangent of the curve at that position.

It is important to avoid this confusion.

clean up TODOs for V4

fix geodesics transformation

The Christoffel symbols do not transform like tensors. This leads to a correction term when dealing with a covariant derivative.

- {\Gamma^\prime}^i_{ab} {u^\prime}^a {u^\prime}^b = \frac{\partial {x^\prime}^i}{\partial x^ j} \left ( -\Gamma^j_{cd} u^c u^d  - \frac{\partial^2 x^j}{\partial x^c \partial {x^\prime}^e} \frac{{x^\prime}^e}{\partial x^d}  u^c u^d\right )

unify naming conventions

• vec/cector/tan/tan_vec/tangent/tangential_vector

remove all todos and turn them into issues

see #51

optimize linalg module

• see POSSIBLE-OPTIMIZATION comments

optimize ray->object intersection test

Calculate the ray once and test against multiple faces/objects at once.

• segmented ray cache
• Runge-Kutta ray cache

feature/optimize-intersection

separate data display from render process

Calculate the (extended) intersection information for each ray first and display it as an image afterwards.
This has the advantage to do post processing and apply filters.

• separate data display
• color filter (normal)
• ray depth filter
• meta data filter (ray step etc.)
• remove old renderers after conversion

fix pylint issue with code duplication in unit tests

• code duplication is allowed in unit tests to keep the dependencies low
• pylint does not support to disable R0801 on a per file basis yet

alternative:

• accept dependencies in unit tests

make miss reason in getter non-optional

is:

IntersectionInfo.miss_reason() -> Optional[MissReason]

should be:

IntersectionInfo.miss_reason() -> MissReason
# precondition: info.misses() == True
# raises RuntimeError

unify demos

• demo submodule?

add support for materials

• u-v maps
• textures
• basic shader

check manifold unittests

• cylindrical

• cylindrical swirl

• invalid value

• edge cases

• non-edge cases

unify manifolds and geometry as much as possible

This should reduce code duplication.

Make geometry an abstraction of its representations as charts.
--> solved by redesign of geometry and introduction of nerte.values.manifolds

pullback/pushforward
--> solved by nerte.values.submanifolds

see also:

• fix/unify-swirl-manifolds
• experimental/transformed_charts

refactor unittests

• intruducde light! dependencies
• investigate usage of built-in assert methods (e.g. assertAlmostEqual)

https://docs.python.org/3/library/unittest.html#type-specific-methods

fix/unittest_v4

add color map to meta info filter

e.g. to highlight specific values like reaching max step

revert code when mypy issue python/mypy/#220 was resovled

see python/mypy#2220

and search for MYPY-BUG in codebase

related: ff10fce

simplify camera detector manifold handling

Handle filter domain application inside camera object?

change camera location to focal point

Define a focal point instead of the camera's location, since the latter is not well defined in arbitrary/curved geometries.

Added alternative projection mode: obscura in 0f1cbf7

Obscura is physically more plausible but might also more difficult to to set up a scene with.

complete cross product

linalg.cross needs completion

requires knowledge of:

• metric
• contra-variant transformation of vectors

extend cylindrical swirl manifold tests

add test cases for swirl != 0

• fixed values for geodesic equation
• coordinate transformation
• vector transformation
• tangential vector transformation
• plane properties

part of #42

add flip normal option to orthographic ray generation

add another ray from pixel mode or add kwargs

bug in intersection test when ray is inside face

add diffeomorphisms between manifolds

Allow to define a manifold based on another manifold.

fix transformation direction for manifolds

The choice of the swirl parameter is currently inconsistent. Make "-a" the flat to swirled transformation and "+a" the inverse transformation.

Let p be a point in space. It is represented by cand s in flat and swirled coordinates respectively.
Let f_a(c) = s, then f_a maps the flat angle phi to the swirled alpha = phi - a *r * z.

The original mistake came from confusing active and passive transformation and assumed alpha = phi + a *r * z.

• cylindrical swirl
• cartesian swirl

part of: #52

integrate color into image filters

optimize cartesian swirl manifold

see POSSIBLE-OPTIMIZATION comments

• Christoffel symbols of first kind are symmetric in two indices. This can be used to reduce the resource usage.

fix typos

-[x] cylindric cylindrical 8913b1a
-[x] carthesian cartesian 3050d30

add rank 3 tensor

add rank3 tensor to linear algebra package

use Runge-Kutta-Nystrom instead of Runge-Kutta

Runge-Kutta-Nystrom may be more efficient than Runge-Kutta for ray propagation.

extend scene functionality

Scene needs more functionality than adding objects, e.g. removing objects.

investigate obscura projection

allow for multi-chart manifolds

add ray step renderer

• add image render filter which displays the amount of ray steps used until a surface was hit similar to the render ray depth mode
• add statistics to support the visual feedback
• requires step count as optional property of intersection info

issue warings instead of runtime errors if possible

rasing an exception unconditionally is not an apropriate way to warn

see: https://docs.python.org/3/library/warnings.html#warnings.warn