AcademySoftwareFoundation / AcademySoftwareFoundation/OpenColorIO

IdentifyBuiltinColorSpace heuristics failure edgecases: ACEScc, ADX16, CIE-XYZ-D65

オープン
#2,222 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る
Feature Request
主要言語
C++
スター
2.1k
フォーク
503
PR マージ指標
30日以内にマージされた PR はありません

説明

Apparently, ACEScc, ADX16, and both CIE XYZ-D65 spaces from ocio://studio-config-latest (as of OCIO v2.5.0) are evading our out-of-the-box matching / identification algorithms.

I've been noticing that, very occasionally, Config::IdentifyBuiltinColorSpace has trouble matching certain color spaces among configs that really do share functionally identical definitions, and should, by all accounts, be recognized as equivalent.

I wrote a little function to assess which color spaces in a given config IdentifyBuiltinColorSpace seems to have the most trouble with -- essentially, I'm using the exact same config for both the srcConfig and the builtinConfig, and just collecting the color spaces that can't be found in their own configs.

In exploring this, I found that both optimization flags (unsurprisingly) and color space visibility (surprisingly) affect the rate of success.

Insights:
- It would be useful if there were a way to force Config.IdentifyBuiltinColorSpace to always attempt to failover to searching inactive color spaces.
- It would be nice if IdentifyBuiltinColorSpace exposed an optional optimizationFlags argument as a means of fine-tuning the equivalency "tolerance"

Findings:
- "ADX16" color spaces based off of the `ADX16_to_ACES2065-1` BuiltinTransform can be matched when using DRAFT optimization or higher
- "ACEScc" and "ACESproxy10i" color spaces based off of the `ACEScc_to_ACES2065-1` `ACESproxy10i_to_ACES2065-1` BuiltinTransform styles respectively remain "unmatchable", regardless of optimization settings.

Here's a python snippet if you wanna mess around:

```python
import PyOpenColorIO as ocio
import os

def get_unmatchable_color_spaces(
config: ocio.Config,
optimization: ocio.OptimizationFlags | int | None = None,
exclude_inactive: bool = False,
verbose: bool = False,
) -> list[str]:

# Optimization Flags affect the identification heuristics. The API doesn't expose a way
# to directly control the optimization flags used for identifying color spaces, but we
# can temporarily adjust the $OCIO_OPTIMIZATION_FLAGS environment variable
# if we want to temporarily override the global optimization behavior

# Preserve the previous environment to restore later.
prev_env = dict(os.environ)
if optimization:
# Temporarily set $OCIO_OPTIMIZATION_FLAGS to the requested value.
os.environ[ocio.OCIO_OPTIMIZATION_FLAGS_ENVVAR] = str(int(optimization))


# Color space visiblity affects the identification heuristics.
# To maximize the likelihood of finding a match, we can temporarily
# activate all color spaces.

# Store the old inactive color spaces to restore later.
old_invisible_color_spaces = config.getInactiveColorSpaces()

if not exclude_inactive:
# Temporarily make all color spaces visible.
config.setInactiveColorSpaces("")


# Iterate over all the color spaces in the config, and collect names of
# color spaces that steadfastly refuse to be matched!
problem_spaces = []
for cs_name in config.getColorSpaceNames(
ocio.SEARCH_REFERENCE_SPACE_ALL,
ocio.COLORSPACE_ALL
):
# Try to match each color space in the config to... itself!
try:
matched_space = ocio.Config.IdentifyBuiltinColorSpace(config, config, cs_name)
except ocio.Exception as e:
if verbose:
ocio.LogMessage(ocio.LOGGING_LEVEL_WARNING, str(e))
problem_spaces.append(cs_name)

# Cleanup -- restore the previous environment and config state.`
os.environ.clear()
os.environ.update(prev_env)
config.setInactiveColorSpaces(old_invisible_color_spaces)

return sorted(problem_spaces)
```

```python
studio_config_modified = ocio.Config.CreateFromFile("ocio://studio-config-latest")

# Add an ACESproxy10i color space to the studio-config, because I suspect it'll also fail to match itself.
studio_config_modified.addColorSpace(
ocio.ColorSpace(
ocio.REFERENCE_SPACE_SCENE,
name="ACESproxy10i",
toReference= ocio.BuiltinTransform("ACESproxy10i_to_ACES2065-1")
)
)

print(f"""
Unmatchable color spaces: {studio_config_modified.getName()}
---

- Default IdentifyBuiltinColorSpace behavior:
{get_unmatchable_color_spaces(studio_config_modified, exclude_inactive=True)}

- Ignore visibility + no optimizations:
{get_unmatchable_color_spaces(studio_config_modified, optimization=ocio.OPTIMIZATION_NONE)}

- Ignore visibility + draft optimizations:
{get_unmatchable_color_spaces(studio_config_modified, optimization=ocio.OPTIMIZATION_DRAFT)}
""")
```

```

Unmatchable color spaces: studio-config-v4.0.0_aces-v2.0_ocio-v2.5
---

- Default IdentifyBuiltinColorSpace behavior:
['ACEScc', 'ACESproxy10i', 'ADX16', 'CIE XYZ-D65 - Display-referred', 'CIE XYZ-D65 - Scene-referred']

- Ignore visibility + no optimizations:
['ACEScc', 'ACESproxy10i', 'ADX16']

- Ignore visibility + draft optimizations:
['ACEScc', 'ACESproxy10i']
```

コントリビューションガイド

コントリビューションガイドを開く

調査の方向性

まず、提供された Python 再現コードを ocio://studio-config-latest に対して実行し、最適化フラグと非アクティブな色空間の設定ごとに IdentifyBuiltinColorSpace の結果を比較します。次に、Config::IdentifyBuiltinColorSpace と、ACEScc、ADX16、CIE XYZ-D65 BuiltinTransform のマッチングパスを調査します。一覧にある同等の色空間が一貫して識別され、要求された可視性と最適化の動作がカバーされれば完了です。

索引モデルが issue の本文から書いたものです。

評価

技術スタック
cpp, python
領域
computer-graphics
issue の種類
バグ
難易度
5/5
見積もり時間
1週間以上
活発さ
停滞
明瞭さ
おおむね明確
初心者へのやさしさ
35/100

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。