fix(sanity): pass if the reference token is empty string

[ktro2828]

What

This pull request introduces a minor adjustment to the reference check logic in base.py. The change ensures that records with an empty string as the reference token are not flagged as missing, improving the accuracy of the sanity check.

• Logic Improvement:
  • Updated the check method in t4_devkit/sanity/reference/base.py to skip records where the reference token is an empty string, treating them as successful matches.

Usecase

If the T4 dataset contains only 2D annotations (object_annotation.json), both first_annotation_token and last_annotation_token will be empty string "", although instance records are generated.

Therefore, fist_annotation_token and last_annotation_token should be the reference to sample_annotation.json.

How PR Tested?

• Parent Link (TIER IV INTERNAL)

• Sample Dataset Link (TIER IV INTERNAL)

• Before

  REF010:
     - No reference to 'instance.first_annotation_token':
     - No reference to 'instance.first_annotation_token':
  REF011:
     - No reference to 'instance.last_annotation_token':
     - No reference to 'instance.last_annotation_token':

+--------------------------------------+---------+--------+--------+---------+----------+
|              DatasetID               | Version | Passed | Failed | Skipped | Warnings |
+--------------------------------------+---------+--------+--------+---------+----------+
| 471a3751-dac2-451e-9421-552975d50272 |         |   47   |   2    |    2    |    4     |
+--------------------------------------+---------+--------+--------+---------+----------+

• After

  REF010: ✅
  REF011: ✅

+--------------------------------------+---------+--------+--------+---------+----------+
|              DatasetID               | Version | Passed | Failed | Skipped | Warnings |
+--------------------------------------+---------+--------+--------+---------+----------+
| 471a3751-dac2-451e-9421-552975d50272 |         |   49   |   0    |    2    |    4     |
+--------------------------------------+---------+--------+--------+---------+----------+

[Copilot]

Pull request overview

This PR fixes the reference validation logic to properly handle empty string reference tokens in T4 datasets that only contain 2D annotations. When a dataset lacks 3D annotations, first_annotation_token and last_annotation_token fields are legitimately empty strings, and should not be flagged as missing references.

Key changes:

• Modified the reference check in base.py to treat empty string reference tokens as valid, preventing false positive validation failures.

---

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

[github-actions]

☂️ Python Coverage

current status: ✅

Overall Coverage

Lines	Covered	Coverage	Threshold	Status
3971	3267	82%	50%	🟢

New Files

No new covered files...

Modified Files

File	Coverage	Status
t4_devkit/sanity/reference/base.py	87%	🟢
TOTAL	87%	🟢

updated for commit: a5e74f5 by action🐍

[ktro2828]

@SamratThapa120 Let me just confirm!
Do you agree with the following idea?

If the T4 dataset contains only 2D annotations (object_annotation.json), both first_annotation_token and last_annotation_token will be empty string "", although instance records are generated.

Therefore, fist_annotation_token and last_annotation_token should be the reference to sample_annotation.json.

As far as I explored, tier4_perception_dataset/AnnotationFileGenerator generate references from an instance record to the first and last annotation records only for 3D annotations, not for 2D.

We can update generator as follows, but it cause conflicts if the dataset contains both 3D and 2D.

Then, I think we should define thafirst_annotation_token and last_annotation_token only refer to sample_annotation records, and they would be empty string "" if the dataset only contains 2D annotations.

[shekharhimanshu]

Thank you for the fix. LGTM! 💯