xiumingzhang / xiuminglib Goto Github PK

Multiple is semantically misleading, e.g., xiuminglib/launch/google.py->Launcher:__gen_borg_file()

E.g., setup_emission_nodetree()

people should be fine with common abbreviations like img and arr.

e.g., util.img can include functions that standardize image shapes (HxW to HxWx1), etc.

Check if functions that should have a writes section have it.

Check if the format is consistent across functions.

read pairs better with write than load.

... so that we can constants.path.checker.

Makes more sense than the current constants.path_checker.

..., since they are not really EXR-related. Maybe start a Normal class and a Depth class?

Old:

the first line
        ... the second line

New:

the first line
... the second line

Make sure all use this style.

Maybe have some overarching util file containing _assert_cycles(), so that modules across the whole package can all use it.

As of now, it's in blender/object.py

Recently I have been trying to run this project of yours
https://github.com/google/neural-light-transport
I am not familiar with Blender, so I have some doubts. I hope you can reply to the question in this project
google/neural-light-transport#4
I hope you can reply
Looking forward to your reply

For instance, config.import_cv2() is weird.

Should use relative imports instead.

from . import config
from .. import os as xm_os

... by renaming it to vis.plot and factoring out those that don't fit into the new name, such as vis.general._save.

Signature: cam.render_buffers(trimesh_obj, alpha=False, depth=False, normals=False, visibility_to=None)

Gradually fix old code

Can you give me some detailed advice on how to customize image composition?
google/neural-light-transport#6
google/neural-light-transport#1
I'm a beginner in neural rendering，I don't quite understand what you mean，I would like you to give a more detailed description or pseudocode
Thank you very much.

Discovered while rendering SLF

Consider moving any non-basic imports into the functions.

This is not inefficient because imports are cached, and more importantly, helps reduce import failures.

If an import is pretty much necessary for all functions in the module (like import bpy to blender.object), wrap the import inside try, so it won't fail at init time (deferring failure till the use time).

preset_import() now supports assert_success. Set it to True for imports that must succeed.

And "X axis" to ":math:x axis"

the image center should be at (h-1)*scale/2 and (w-1)*scale/2. It seems the intrinsics from here and here is off by one pixel.

Changing it to the correct offset gives correct depth reprojection.

Move depth/normal/etc. visualization to either vis or geometry, and switch to tfg.io.exr for easier OpenEXR installation (?).

and make_apng calls it

See nerfactor/data_gen/dtu_mvs/surf_from_mvs.py on how I extracted K from P with cv2 but then later found [R|t] with K^{-1}P, instead of using those from the cv2 decomposition (which uses another weird convention).