Restore RenderEngineVtk MultiMaterialObject test for EGL #22053
Conversation
The crux of this was that rainbow_stripes.png has an alpha channel. When
we enabled shadows in VTK, we swapped out rendering pipelines which treated
meshes with those images differently. The test never really noticed that
these changed. Changing the backend merely caused the change to manifest
in a way that the test noticed (merely by happenstance).
The solution:
- Remove the alpha channel.
- Modify the test so it will be more sensitive to those kinds of changes.
- Refactor the CompareImage code to use more widely (removing its
behavior of saving the image to compare).
- Make use of VTK's automatically save rendered image functionality.
- Update callsites to CompareImage accordingly.
This does mean we don't have a test of transparent things. This is not a
problem:
1. We only accidentally had one before.
2. We don't support transparency with any kind of uniformity or
reliability.
3. We need a test that tracks what we *do* support and document
supported transparency appropriately (a follow up PR).
SeanCurtis-TRI
force-pushed
the
PR_fix_multi_mat_test
branch
from
3193f8d
to
faf218e
Compare
|
Please choose a feature reviewer if this is ready, or switch to Draft. |
There was a problem hiding this comment.
+@jwnimmer-tri for feature review.
Reviewable status: LGTM missing from assignee jwnimmer-tri(platform), needs at least two assigned reviewers, missing label for release notes (waiting on @SeanCurtis-TRI)
SeanCurtis-TRI
added
the
release notes: none
There was a problem hiding this comment.
+(release notes: none)
Reviewable status: LGTM missing from assignee jwnimmer-tri(platform), needs at least two assigned reviewers
There was a problem hiding this comment.
Reviewed 4 of 4 files at r1, all commit messages.
Reviewable status: 3 unresolved discussions, needs at least two assigned reviewers
geometry/render_vtk/test/internal_render_engine_vtk_test.cc line 309 at r1 (raw file):
// located at `ref_filename` to within the given tolerance. template <typename ImageType> void CompareImages(const ImageType& test_image, const std::string& ref_filename,
nit This is a resource path, not a filename. Maybe "ref_resource" or "ref_resource_path" or "ref_respath" or something?
Code quote:
ref_filenamegeometry/render_vtk/test/internal_render_engine_vtk_test.cc line 996 at r1 (raw file):
"drake/geometry/render/test/meshes/rainbow_box.obj"))}; // The camera has a smaller image (640 x 480), a narrower field of view (so
typo
Suggestion:
480 x 360geometry/render_vtk/test/internal_render_engine_vtk_test.cc line 1040 at r1 (raw file):
// PBR. OBJs get rendered using Phong materials. It leads to slightly // different levels of illumination, requiring a bit more tolerance (about // 10%). When looking at the images, they should be qualitatively similar.
BTW Another way to do this would be to use two different goal images (both committed into git), so that tolerance is tight in both cases. Maybe that's better? You can decide.
The crux of this was that rainbow_stripes.png has an alpha channel. When we enabled shadows in VTK, we swapped out rendering pipelines which treated meshes with those images differently. The test never really noticed that these changed. Changing the backend merely caused the change to manifest in a way that the test noticed (merely by happenstance).
The solution:
This does mean we don't have a test of transparent things. This is not a problem:
Resolves #22044
This change is