Store intensity config in IR metadata by kprokofi · Pull Request #6059 · open-edge-platform/training_extensions

kprokofi · 2026-04-09T06:17:07Z

Summary

Resolves #5952

How to test

Checklist

The PR title and description are clear and descriptive
I have manually tested the changes
All changes are covered by automated tests
All related issues are linked to this PR (if applicable)
Documentation has been updated (if applicable)

github-actions · 2026-04-09T06:22:44Z

🐳 Docker image sizes

Device	Size
`cpu`	3592.9 MB (3.51 GB)
`xpu`	10561.5 MB (10.31 GB)
`cuda`	11713.8 MB (11.44 GB)

Copilot

Pull request overview

This PR implements exporting of IntensityConfig (high-bit-depth / domain-specific intensity preprocessing) into exported model metadata (rt_info / ONNX metadata), by propagating intensity configuration from the data pipeline through CLI/Engine into DataInputParams, and finally embedding it during export. This addresses the need from issue #5952 to make exported models self-describing for inference-time preprocessing reconstruction.

Changes:

Extend DataInputParams to optionally carry IntensityConfig and serialize it via as_dict().
Plumb intensity config from OTXDataModule → CLI/Engine model instantiation → exporter metadata embedding.
Add unit tests covering metadata embedding for multiple IntensityConfig modes and DataInputParams.as_dict() behavior.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
`library/src/otx/backend/native/exporter/base.py`	Embed intensity-related keys (dtype, mode, parameters) into export metadata.
`library/src/otx/backend/native/models/base.py`	Add `intensity_config` to `DataInputParams`; parse intensity config dicts in preprocessing params.
`library/src/otx/data/module.py`	Expose training subset intensity config on the datamodule for downstream plumbing.
`library/src/otx/cli/cli.py`	Forward datamodule intensity config into `data_input_params` during CLI instantiation.
`library/src/otx/backend/native/engine.py`	Forward datamodule intensity config into model construction args.
`library/tests/unit/backend/native/exporter/test_base.py`	Add tests for intensity metadata embedding and `DataInputParams.as_dict()`.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-09T06:36:47Z

+        # Forward the training subset's intensity config so that Engine / CLI
+        # can pass it to OTXModel.data_input_params → exporter rt_info.
+        # None means no subset or no intensity attr (old configs / no data pipeline).
+        self.input_intensity_config = getattr(self.train_subset, "intensity", None)
+


OTXDataModule.from_otx_datasets() does not set input_intensity_config, but Engine/CLI now read datamodule.input_intensity_config. If a datamodule is created via from_otx_datasets (e.g. quantization/AutoConfigurator path), this can raise AttributeError. Consider initializing input_intensity_config in from_otx_datasets as well (e.g. from train_subset.intensity) or ensuring callers use getattr(..., None).

Copilot · 2026-04-09T06:36:47Z

+            if self._datamodule.input_intensity_config is not None:
+                params["intensity_config"] = self._datamodule.input_intensity_config


Directly accessing self._datamodule.input_intensity_config can raise AttributeError for datamodules created via OTXDataModule.from_otx_datasets(), which currently doesn’t define that attribute. Either ensure the datamodule factory sets input_intensity_config or use getattr(self._datamodule, "input_intensity_config", None) here.

Suggested change

if self._datamodule.input_intensity_config is not None:

params["intensity_config"] = self._datamodule.input_intensity_config

input_intensity_config = getattr(self._datamodule, "input_intensity_config", None)

if input_intensity_config is not None:

params["intensity_config"] = input_intensity_config

Copilot · 2026-04-09T06:36:48Z

+            if self.datamodule.input_intensity_config is not None:
+                _dip["intensity_config"] = self.datamodule.input_intensity_config


self.datamodule.input_intensity_config may not exist for datamodules created via OTXDataModule.from_otx_datasets() (it currently doesn’t define that attribute). This can raise AttributeError during CLI instantiation; consider using getattr(self.datamodule, "input_intensity_config", None) or making from_otx_datasets initialize it.

Suggested change

if self.datamodule.input_intensity_config is not None:

_dip["intensity_config"] = self.datamodule.input_intensity_config

input_intensity_config = getattr(self.datamodule, "input_intensity_config", None)

if input_intensity_config is not None:

_dip["intensity_config"] = input_intensity_config

Copilot · 2026-04-09T06:36:48Z

+            # Map storage_dtype to the ModelAPI input_dtype convention
+            _dtype_map = {"uint8": "u8", "uint16": "u16", "float32": "f32"}
+            input_dtype = _dtype_map.get(intensity_cfg.storage_dtype, intensity_cfg.storage_dtype)


IntensityConfig.storage_dtype supports values beyond {uint8,uint16,float32} (e.g. "int16" per otx/config/data.py). The dtype mapping here will fall back to the raw string, which may not match the intended ModelAPI input_dtype convention. Consider extending _dtype_map (and/or validating storage_dtype) to cover all supported dtypes consistently.

Suggested change

# Map storage_dtype to the ModelAPI input_dtype convention

_dtype_map = {"uint8": "u8", "uint16": "u16", "float32": "f32"}

input_dtype = _dtype_map.get(intensity_cfg.storage_dtype, intensity_cfg.storage_dtype)

# Map storage_dtype to the ModelAPI input_dtype convention.

# Keep the original value for unknown dtypes to preserve current behavior,

# but warn because ModelAPI expects the shorthand form (for example, "i16").

_dtype_map = {

"int8": "i8",

"int16": "i16",

"int32": "i32",

"uint8": "u8",

"uint16": "u16",

"uint32": "u32",

"float16": "f16",

"float32": "f32",

"float64": "f64",

}

storage_dtype = intensity_cfg.storage_dtype

normalized_storage_dtype = storage_dtype.lower()

input_dtype = _dtype_map.get(normalized_storage_dtype, storage_dtype)

if input_dtype == storage_dtype and normalized_storage_dtype not in _dtype_map:

log.warning(

"Unsupported intensity storage_dtype '%s' for ModelAPI input_dtype mapping; "

"using raw value in exported metadata.",

storage_dtype,

)

Copilot · 2026-04-09T06:36:48Z

+            extra_data[("model_info", "intensity_mode")] = intensity_cfg.mode
+
+            if intensity_cfg.max_value is not None:
+                extra_data[("model_info", "intensity_max_value")] = str(intensity_cfg.max_value)
+            if intensity_cfg.window_center is not None:


IntensityConfig.max_value has an 'auto' meaning when None (see otx/config/data.py and build_intensity_transform auto-detection). For mode="scale_to_unit", the exporter currently omits intensity_max_value when max_value is None, so the exported rt_info can’t fully represent the effective preprocessing parameters unless the consumer duplicates the same auto-logic. Consider resolving and exporting the effective max_value (at least for non-uint8 dtypes) to make exports self-contained.

codecov-commenter · 2026-04-09T10:57:21Z

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 95.00000% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
library/src/otx/backend/native/models/base.py	77.77%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

kprokofi added 7 commits March 31, 2026 04:57

fix augmentation order, rfdetr OV and input size from_otx_dataloader

a4c1c63

reply comments

ffaecce

reply comments

7d3fcac

fix semantic seg

d7f0081

fix aspect ratio

106f48e

minor fix

6effdb0

add additional meta data

017b530

github-actions bot added the TEST Any changes in tests label Apr 9, 2026

Merge branch 'develop' into kprokofi/introduce-ir-meta-data

d318c20

kprokofi marked this pull request as ready for review April 9, 2026 06:28

kprokofi requested a review from a team as a code owner April 9, 2026 06:28

Copilot AI review requested due to automatic review settings April 9, 2026 06:28

Copilot started reviewing on behalf of kprokofi April 9, 2026 06:29 View session

Copilot AI reviewed Apr 9, 2026

View reviewed changes

reply comments, fix test

6ba80db

kprokofi added 2 commits April 10, 2026 17:25

minor fix of ruff:

20439d6

Merge branch 'develop' into kprokofi/introduce-ir-meta-data

427303a

leoll2 previously approved these changes Apr 10, 2026

View reviewed changes

fix integration test

b76dd27

kprokofi dismissed leoll2’s stale review via b76dd27 April 10, 2026 12:40

leoll2 changed the title ~~Store Intencity Config in IR meta data~~ Store intensity config in IR metadata Apr 11, 2026

kprokofi added 2 commits April 13, 2026 18:00

fix ruff

3ce1d65

merge develop

9a27b81

leoll2 reviewed Apr 13, 2026

View reviewed changes

Comment thread library/src/otx/backend/native/exporter/base.py Outdated

reply comment?

7bf9b69

kprokofi requested a review from leoll2 April 13, 2026 13:14

leoll2 approved these changes Apr 13, 2026

View reviewed changes

kprokofi enabled auto-merge April 13, 2026 13:56

kprokofi added this pull request to the merge queue Apr 13, 2026

Merged via the queue into develop with commit 0590444 Apr 13, 2026
49 checks passed

kprokofi deleted the kprokofi/introduce-ir-meta-data branch April 13, 2026 15:10

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Store intensity config in IR metadata#6059

Store intensity config in IR metadata#6059
kprokofi merged 15 commits intodevelopfrom
kprokofi/introduce-ir-meta-data

kprokofi commented Apr 9, 2026

Uh oh!

github-actions bot commented Apr 9, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

codecov-commenter commented Apr 9, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

		if self._datamodule.input_intensity_config is not None:
		params["intensity_config"] = self._datamodule.input_intensity_config

-            if self._datamodule.input_intensity_config is not None:
-                params["intensity_config"] = self._datamodule.input_intensity_config
+            input_intensity_config = getattr(self._datamodule, "input_intensity_config", None)
+            if input_intensity_config is not None:
+                params["intensity_config"] = input_intensity_config

		if self.datamodule.input_intensity_config is not None:
		_dip["intensity_config"] = self.datamodule.input_intensity_config

-            # Map storage_dtype to the ModelAPI input_dtype convention
-            _dtype_map = {"uint8": "u8", "uint16": "u16", "float32": "f32"}
-            input_dtype = _dtype_map.get(intensity_cfg.storage_dtype, intensity_cfg.storage_dtype)
+            # Map storage_dtype to the ModelAPI input_dtype convention.
+            # Keep the original value for unknown dtypes to preserve current behavior,
+            # but warn because ModelAPI expects the shorthand form (for example, "i16").
+            _dtype_map = {
+                "int8": "i8",
+                "int16": "i16",
+                "int32": "i32",
+                "uint8": "u8",
+                "uint16": "u16",
+                "uint32": "u32",
+                "float16": "f16",
+                "float32": "f32",
+                "float64": "f64",
+            }
+            storage_dtype = intensity_cfg.storage_dtype
+            normalized_storage_dtype = storage_dtype.lower()
+            input_dtype = _dtype_map.get(normalized_storage_dtype, storage_dtype)
+            if input_dtype == storage_dtype and normalized_storage_dtype not in _dtype_map:
+                log.warning(
+                    "Unsupported intensity storage_dtype '%s' for ModelAPI input_dtype mapping; "
+                    "using raw value in exported metadata.",
+                    storage_dtype,
+                )

Conversation

kprokofi commented Apr 9, 2026

Summary

How to test

Checklist

Uh oh!

github-actions bot commented Apr 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🐳 Docker image sizes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

codecov-commenter commented Apr 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

github-actions bot commented Apr 9, 2026 •

edited

Loading

codecov-commenter commented Apr 9, 2026 •

edited

Loading