Some files like FBX support lights and cameras, but these are not always useful.

Assimp supports an importer post-process called aiProcess_RemoveComponent and configured with:

importer.SetPropertyInteger(AI_CONFIG_PP_RVC_FLAGS, aiComponent_CAMERAS | aiComponent_LIGHTS);

We might want yoga to remove everything but meshes/materials by default, and provide a --no-cleanup option in the CLI and such.

Draco is a powerful 3D library to compress 3D meshs: https://github.com/google/draco

What about Yoga compatibility? it could be a must to have.

Someone requested an option to keep metadata, so I open this issue to keep the information.

Questions that should be answered before:

• How to achieve this for each formats?
• What to do when the input format is different than the output format (converting metadata? What can be kept ?)...

Or if applied to different meshes.

See @nekohayo issue on YOGA Image Optimizer → flozz/yoga-image-optimizer#20

Hello,

Because of the very long computation time for a lot of images (#28), a progression indicator (like a progress bar, or just a vague percentage) would be useful to estimate how long the processing will take.
What do you think about such a feature ? Any chance it gets implemented ?

Thank you

I used yoga to convert a fbx (composed of two submeshes) into a glb and the normals of one of submesh are inverted.

There is a screen:

I set Blender in display normals mode. Blue color is normals facing the camera

How the ArchLinux package is compiled: Depend on assimp, because unable to install it following instruction. The assimp source code is still needed to compile yoga. So

• get assimp-5.0.1.tar.gz
• remove assimp/build
• untar assimp and add link at the same place
  python3 setup.py build_py

build_py is the only one that don't make error, and need this lib. The other documented build process don't work on Arch.
The files are then placed in the package dir using:
python setup.py install --root="${pkgdir}" --optimize=1 --skip-build

After installing the made package, it works fine with the user that build the package, but with other user, there is strangely the following error. could be due to some parts that are installed only for the compiling user during the build process?

Traceback (most recent call last):
  File "/usr/bin/yoga", line 33, in <module>
    sys.exit(load_entry_point('yoga==1.0.0', 'console_scripts', 'yoga')())
  File "/usr/bin/yoga", line 25, in importlib_load_entry_point
    return next(matches).load()
  File "/usr/lib/python3.9/importlib/metadata.py", line 77, in load
    module = import_module(match.group('module'))
  File "/usr/lib/python3.9/importlib/__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 972, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 986, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 680, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 855, in exec_module
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "/usr/lib/python3.9/site-packages/yoga/__init__.py", line 2, in <module>
    from . import model  # noqa
  File "/usr/lib/python3.9/site-packages/yoga/model/__init__.py", line 174, in <module>
    from .assimp import assimp_import_from_bytes, assimp_export_to_bytes
  File "/usr/lib/python3.9/site-packages/yoga/model/assimp.py", line 6, in <module>
    from ._assimp import lib, ffi
ModuleNotFoundError: No module named 'yoga.model._assimp'

Currently, yoga is not able to import a glb, as the check for the format wants to make sure that the filename ends with .glb (and it does not because assimp reads memory).

Provide a pHint argument to readFileFromMemory: http://assimp.sourceforge.net/lib_html/class_assimp_1_1_importer.html#a9b3c5e8b1042702f449e84a95b3324f6

Generate it from lowercased the filename's extension (if any).

Expected Behavior

When trying to optimize a glb file, application should return an optimized glb file.

Current Behavior

When trying to optimize a glb file, application returns ValueError: Cannot resolve *13 and quits

Steps to Reproduce

1. Using latest Windows version - yoga 1.3.1
2. Using latest Windows version of Blender - blender 4.0.2
3. Using Blender glb/gltf exporter, export 3D model
4. Using Yoga, run yoga model input.glb output.glb

Context (Environment)

• OS: Win 11
• Blender: 4.0.2
• Yoga: 1.3.1 / 1.3.0

Detailed Error Log

Traceback (most recent call last):
  File "C:\yoga\yogawin.py", line 6, in <module>
  File "C:\yoga\yoga\__main__.py", line 25, in main
  File "C:\yoga\concurrent\futures\_base.py", line 449, in result
  File "C:\yoga\concurrent\futures\_base.py", line 401, in __get_result
  File "C:\yoga\concurrent\futures\thread.py", line 58, in run
  File "C:\yoga\yoga\model\__init__.py", line 251, in optimize
  File "C:\yoga\yoga\model\helpers.py", line 135, in model_embed_images
ValueError: Cannot resolve *13

Possible solution

Maybe is something related to the blender exported file?

Summary

As a user I want to be able to keep Exif/XMP/IPTC or other metadata when optimizing an image, so that metadata affecting image rendering are kept.

Why

As we've discussed on discord I've found that when optimizing a file, yoga strip it from metadata. It result in a change of color of resulting image, weither when optimizing to webp or jpg.

Example file

Below are source file in JPG, and webp image in a zip (github is not compatible with webp upload)
img_webp.zip

img_brute_photoshop.jpg

Raw image from a photoshop export, lot of EXIF metadata

img_export_gimp_no_exif.webp

Converted image using Gimp, there is still metadata as seen with this extract:

+-- img_export_gimp_no_exif.webp
    +-- RIFF [size: 16697518, formtype: WEBP]
        +-- VP8X [offset: 20, size: 10]
            +-- has_icc: 1
            +-- has_alpha: 0
            +-- has_exif: 1
            +-- has_xmp: 0
            +-- has_anim: 0
            +-- canvas_width: 4252
            +-- canvas_height: 2835
        +-- ICCP [offset: 38, size: 532]
        +-- VP8L [offset: 578, size: 16696926]
        +-- EXIF [offset: 16697512, size: 14]

img_optimized_noresized.webp

Optimised image without resizing, with a loss of color, no metadata.

img_optimized_resized.webp

Optimised image with resizing, same loss of color, no metadata.

See flozz/yoga-image-optimizer#14 (comment)

The documentation says:

Building YOGA on Windows is not supported yet...

It could be nice to be able to use YOGA also on Windows!

Add inside documentation that python-setuptools build-essential and python-dev are needed

Hello,
Any chance that Yoga includes multi-threading in the future ?

Overall performance seems pretty ok for small jpeg images, but for bigger ones it's really slow. One 75MB PNG (50Mpixels) took 30min to be converted, a 1.6MB PNG took 6min (but 33% saved, nice :).

Thanks

While embedding textures, an image referenced as ../../image.png,
but present in root_path directly won't be found.

• The file is set to an empty one (zero bytes) even if it fails.

Try :

• Load a model with a missing texture, it will erase a eventually previous output file.

Hello,

I did convert a few files and notices a very high memory usage (often about 12 times the size of the original png), and in particular my last test with a 9000x3000 PNG file (44MB) did use more than 20GB of RAM, which seems extreme for such a file…
So I wonder if that's something expected (and if yes, what would be possible to reduce the memory usage ?) or a bug on my side.

Thanks in advance

Have a dict:

{
    "texturePath0": <FileLike>,
    "texturePath1": <FileLike>,
}

This is a bit of a corner case, but it is useful for massively shrinking size further when dealing with scanned documents, especially text or some types of drawings: the ability to force the image to be in grayscale instead of RGB color. Oftentimes, document scanning apps (or people who do the scanning) may not think (or know) to encode that way, so the image is imperceptibly color even though perceptually it is grayscale. Converting to grayscale, for PNG and other lossless (or lossy) formats helps a ton... but doing it one image at a time with GIMP is a pain in the butt.

The same rule applies to alpha channels: sometimes there is one even if you don't necessarily need it, and it's extra weight that could be discarded (if the user knows what they're doing :)

Hello,

I would like to know if you are planning to implement a Graphical User Interface for that software ?

Thanks

Fill the texture with some kind of a debug image.

Looking at zeux/meshoptimizer, we could use some optimization to assimp output geometries.

First do some tests with outputted geometries by YOGA, count the gains, and see if an integration as a lib is doable.

TODO-list:

• Test gltf optimizer with multiple geometries, measure the gains if any
• See if it is usable as a library to add to YOGA (python bindings)
• Write python bindings
• Integrate to YOGA (and option to deactivate those optimizations ?)

Just installed Yoga and I cannot browse /var/www to open files in my local web dev directories. Is there a way for me to let Yoga see these directories? I have tried via 'Other locations' with no success.

In the installation doc, the instructions to get the git submodule are missing

In yoga/model/assimp.cpp, extract_image_nodes does not extract already embedded textures, they are ignored.

Keeping the very same convention than assimp, we could just fill imageNode with:

auto imageNode = new ImageNode();
imageNode->path = "*1"; // Same convention, the 1 is the index of the textures in scene->mTextures
imageNode->bytes = /* extract bytes for scene->mTextures */;
imageNode->bytes_length = /* extract bytes length */;

I'v tried to convert this model
artist.zip

And yoga raised this error:

UnicodeDecodeError: 'ascii' codec can't decode byte 0xc3 in position 11: ordinal not in range(128)

Add choices for JPEG optimization:

• Guetzli + MozJPEG

  • Always reencode JPEGs using Guetzli and then use additional MozJPEG (lossless) optimization

• MozJPEG

  • Always reencode JPEGs using libjpeg and then optimize losslessly with MozJPEG

• MozJPEG lossless

  • If the input image is already a JPEG: only perform a lossless optimization using MozJPEG (ignore the quality parameter)
  • If the input image is not a JPEG: encode the image using libjpeg (Pillow) and then optimize it with MozJPEG (lossless)

Related issue: flozz/yoga-image-optimizer#15

When doing a JPEG to JPEG optimization, we should never have an output image bigger than the input one.

• If the guetzli+mozjpeg reencoding produce a smaller image → keep it
• else, try a mozjpeg (lossless) optimization alone, and if it produces a smaller image → keep this one
• else, just remove metadata and keep the original encoding.

Issue #43 must be resolved before this one

Related issue: flozz/yoga-image-optimizer#14

There is no documentation on about how to build yoga and its dependencies.

FYI @flozz