You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
docs(skill): expand anomalib-adding-a-datamodule with a few missed points - #3830
…#3830
- Clarify that _create_val_split() is mode-driven (not sample-driven
like _create_test_split()); FROM_DIR requires self.val_data to be
set manually in _setup().
- Use a HEAD request with redirect-following for the gating-check
curl snippet, instead of a plain GET that transfers the full file.
- Generalize the docs-registration step to cover image/video/depth
reference folders instead of hardcoding image/.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This prescribes only an unrecorded manual check for the exact non-standard-format failure that the skill identifies, so a generated datamodule PR can still have no repeatable regression coverage. Require a deterministic fixture that writes representative bit depths/channel layouts and asserts the result through read_image(); real-sample inspection can remain supplemental.
This issue also appears on line 334 of the same file.
…#3830
- Clarify that _create_val_split() is mode-driven (not sample-driven
like _create_test_split()); FROM_DIR requires self.val_data to be
set manually in _setup().
- Use a HEAD request with redirect-following for the gating-check
curl snippet, instead of a plain GET that transfers the full file.
- Generalize the docs-registration step to cover image/video/depth
reference folders instead of hardcoding image/.
… review
Add guidance gaps surfaced while reviewing PR open-edge-platform#3800 (LIBAD dataset):
- prepare_data() download patterns (open vs gated datasets, with
BMAD/Kaputt references)
- Reuse AnomalibDataModule._create_test_split()/_create_val_split()
instead of hand-rolled splitting
- Validate dataset-specific params (category/modality) with clear errors
- Verify License/Reference docstring accuracy against the dataset's
actual source, not the paper's code repo
- Docs page + examples config + CHANGELOG registration steps
- Sanity-check real pixel data for non-standard image formats, since
dummy test generators only produce 8-bit images
- Updated reviewer checklist to match
Signed-off-by: Rajesh Gangireddy <rajesh.gangireddy@intel.com>
…#3830
- Clarify that _create_val_split() is mode-driven (not sample-driven
like _create_test_split()); FROM_DIR requires self.val_data to be
set manually in _setup().
- Use a HEAD request with redirect-following for the gating-check
curl snippet, instead of a plain GET that transfers the full file.
- Generalize the docs-registration step to cover image/video/depth
reference folders instead of hardcoding image/.
Signed-off-by: Rajesh Gangireddy <rajesh.gangireddy@intel.com>
Drop mid-sentence/clause-level bold that doesn't match the skill's
existing convention of using bold only as a short label (e.g. 'Open
dataset:', 'Never trust dataset paths:'). Code spans already carry
enough emphasis without wrapping full sentences in bold.
Signed-off-by: Rajesh Gangireddy <rajesh.gangireddy@intel.com>
This unconditional requirement does not apply to generic/internal loaders: the existing Folder and Tabular dataset/datamodule pairs have no external dataset license or paper to cite, and some third-party datasets may have no arXiv ID or DOI. Qualify the rule to published third-party datasets to avoid prompting contributors to invent inapplicable metadata.
Document re-exporting datasets through the public API
The registration flow stops after exporting the datamodule, but source-specific dataset classes are also consistently re-exported through src/anomalib/data/datasets/image/__init__.py:30-44, src/anomalib/data/datasets/__init__.py:46-58, and src/anomalib/data/__init__.py:75-89. Please add these steps so following this skill does not leave MyDataset outside the established public API.
test_split_ratio is set describes a state the base class does not preserve: AnomalibDataModule.__init__ converts None (and other falsey values) to 0.5. This could lead contributors to assume that passing None suppresses sampling. Document the actual trigger and effective default instead.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Based on recent PRs , I think these additions would make it much more clear in creating a more complete "adding datasets" PR.
📝 Description
✨ Changes
Select what type of change your PR is:
✅ Checklist
Before you submit your pull request, please make sure you have completed the following steps:
For more information about code review checklists, see the Code Review Checklist.