🆕 Add Cerberus#1060
Draft
measty wants to merge 101 commits into
Draft
Conversation
for more information, see https://pre-commit.ci
- Drop Python 3.10 support - Add Python 3.14 support
for more information, see https://pre-commit.ci
…remove-python3.10
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## develop #1060 +/- ##
===========================================
- Coverage 99.88% 99.63% -0.25%
===========================================
Files 85 93 +8
Lines 11623 12166 +543
Branches 1524 1598 +74
===========================================
+ Hits 11610 12122 +512
- Misses 7 20 +13
- Partials 6 24 +18 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
shaneahmed
reviewed
May 22, 2026
|
|
||
|
|
||
| class Cerberus(ModelABC, NetDesc): | ||
| """Cerberus multi-task model for glands, lumen, nuclei, and patch class.""" |
Member
There was a problem hiding this comment.
Can you expand the description and add some metrics reported in the paper as in the link below?
https://tia-toolbox.readthedocs.io/en/latest/_autosummary/tiatoolbox.models.architecture.kongnet.KongNet.html
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
Comment on lines
+39
to
+45
| def __init__( | ||
| self, | ||
| patch_output_shape: tuple[int, int] = (144, 144), | ||
| nuclei_type_dict: dict | None = None, | ||
| gland_type_dict: dict | None = None, | ||
| lumen_type_dict: dict | None = None, | ||
| ) -> None: |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR introduces the Cerberus multi-task (gland/lumen/nuclei + patch-class) architecture and wires it into TIAToolbox’s pretrained-model registry, alongside engine improvements to support halo-based tile post-processing and some utility additions.
Changes:
- Added Cerberus architecture (network, backbone, post-processing) + registry entry and unit tests.
- Enhanced
MultiTaskSegmentortile-mode post-processing with an optionalpostproc_haloto improve boundary handling and object ownership across tiles. - Added an annotation-store utility (
combine_annotation_stores) with tests, and improved weight-loading compatibility for checkpoints wrapped under"desc"/DataParallelprefixes.
Reviewed changes
Copilot reviewed 19 out of 19 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
tiatoolbox/wsicore/wsireader.py |
Adjusts tissue masking to neutralize pure-black pixels in thumbnails before masking. |
tiatoolbox/models/models_abc.py |
Makes load_torch_model robust to "desc"-wrapped checkpoints and module.-prefixed keys. |
tiatoolbox/models/engine/multi_task_segmentor.py |
Adds postproc_halo support for tile post-processing + improves tile merging/spill-to-zarr behavior. |
tiatoolbox/models/engine/io_config.py |
Extends IO config schema/docs to carry postproc_halo. |
tiatoolbox/models/architecture/cerberus/** |
Adds Cerberus model definition, minimal ResNet-34 encoder, decoder utilities, and post-processing. |
tiatoolbox/models/__init__.py |
Exposes Cerberus at tiatoolbox.models.Cerberus. |
tiatoolbox/data/pretrained_model.yaml |
Registers cerberus-resnet34 pretrained entry (HF repo + IO config). |
tiatoolbox/annotation/utils.py |
Adds combine_annotation_stores helper. |
tiatoolbox/annotation/__init__.py |
Re-exports combine_annotation_stores. |
tests/test_annotation_utils.py |
Tests for combine_annotation_stores. |
tests/models/test_arch_cerberus.py |
Unit tests for Cerberus model, postproc, and pretrained registry wiring. |
tests/engines/test_multi_task_segmentor.py |
Tests for halo post-processing and vertical merge after Zarr spill. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+1789
to
+1790
| # set any black pixels to white to avoid them being included in the tissue mask | ||
| thumbnail[thumbnail.sum(axis=2) == 0] = 255 |
Comment on lines
+56
to
+74
| labels_ = _normalise_labels(input_path_objs, labels) | ||
| combined_store = SQLiteStore(auto_commit=False) | ||
|
|
||
| for source_path in input_path_objs: | ||
| source_store = SQLiteStore.open(source_path) | ||
| source_label = labels_[source_path] | ||
| annotations = [] | ||
| keys = [] | ||
| for key, annotation in source_store.items(): | ||
| properties = dict(annotation.properties) | ||
| properties[label_property] = source_label | ||
| annotations.append(Annotation(annotation.geometry, properties)) | ||
| keys.append(f"{source_label}:{key}") | ||
| if annotations: | ||
| combined_store.append_many(annotations, keys) | ||
|
|
||
| combined_store.commit() | ||
| combined_store.dump(output_path) | ||
| return output_path |
Comment on lines
+127
to
+144
| def _dilate_labelled_instances(inst_fg: np.ndarray, k_disk: np.ndarray) -> np.ndarray: | ||
| """Label foreground instances, dilate each object, and fill holes.""" | ||
| inst_lab = label(inst_fg)[0] | ||
| output_map = np.zeros(inst_lab.shape) | ||
| for inst_id in np.unique(inst_lab).tolist()[1:]: | ||
| inst_map = np.array(inst_lab == inst_id, dtype=np.uint8) | ||
| y1, y2, x1, x2 = get_bounding_box(inst_map) | ||
| pad = k_disk.shape[0] * 2 | ||
| y1 = max(y1 - pad, 0) | ||
| x1 = max(x1 - pad, 0) | ||
| x2 = min(x2 + pad, inst_map.shape[1] - 1) | ||
| y2 = min(y2 + pad, inst_map.shape[0] - 1) | ||
| inst_map_crop = inst_map[y1:y2, x1:x2] | ||
| inst_map_crop = cv2.dilate(inst_map_crop, k_disk, iterations=1) | ||
| inst_map_crop = binary_fill_holes(inst_map_crop) | ||
| output_region = output_map[y1:y2, x1:x2] | ||
| output_region[inst_map_crop > 0] = inst_id | ||
| return output_map |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This PR adds Cerberus model architecture and pretrained models, so it can be used within the newly-efficient tiatoolbox engines.
Still a draft as needs tidying up a bit and uploading weights to huggingface.
Example output:
