Feature/rotate cropped image #82

jokkalt · 2026-01-08T16:04:21Z

Implements the following flow from the matlab files RotateCropImage.m and RemoveHolesAndStitchesNew.m:
- Determine the rotation angle for the image and mask
By using the given rotation_angle if rotation_angle != 0. Otherwise, if the first object of crop_info is a
rectangle, then the rotation_angle of that rectangle is determined. Any other case will lead to a rotation angle
of 0.
- If the rotation angle is not 0, the mask is binary dilated using DILATE_STEPS iterations to correct for
imperfections when rotating. A margin is determined to reduce the final image to compensate for the dilation.
- The mask and image are cropped to the bounds of the mask.
- The scan image is cleaned of needles. The parameter times_median is used to determine the threshold to find
outliers. (this step is RemoveHolesAndStitchesNew.m; note that the parameter interpolate_data in that script is ALWAYS 0, so a lot of code is skipped)
- The cleaned image is masked using the cropped mask.
- The image and mask are rotated by rotation_angle. (This code is freshly written, not using matlab code, since matlab code did not translate well due to java/matlab/python coordinate system used)
- The rotated image and rotated mask are cropped to the bounds of the mask, using margin to compensate for dilation.

NOTE: there was a bug in the existing crop_to_mask function: y and x were switched

Returns will be use to wrap function in container that can be passed around in rail. Thus this will be use for railway function methodology

We no longer are working with the raw value, since it's unclear at runtime if a failure occured or not. Thus the raw value will be encapsulated in a container that can be passed around until the entended reciever consumes it; think of a mail currier package. State of the container can be Success or Failure, but that can be unpack lazily

Writen module docstrings for new modules with the assistant of Claud (AI)

Pipelines is an abstracted module that communicates with the returns layer. This means endpoints can call scratch-core function without caring without caring what the underlying protocol is used for failures/exceptions

Create an upload scan parameters model that hold all the parameters to be used for the pipelines. All the parameters have a defaults and defaults will be generated if no parameters is given, thus nothing changes for the user or endpoint contract

deptry notify that dependencies needed maintainance

* mask and crop functionality added * type fixes and some efficiency

gaussian filter --------- Co-authored-by: Sharlon N. Regales <[email protected]>

* resampling method added

* Mark type and resampling for marks implemented

added files to lfs

SimoneAriens · 2026-01-19T10:16:08Z

packages/scratch-core/src/conversion/data_formats.py

+    selected area.
+
+    The points dict differs per CropType:
+    CIRCLE: {'center': array [x, y], 'radius': float}


is there any way we could enforce/test/validate this when creating a CropInfo?

cfs-data · 2026-01-19T14:44:38Z

packages/scratch-core/src/conversion/data_formats.py

+    ELLIPSE: {'center': array [x, y], 'majoraxis': float, 'minoraxis': float, angle_majoraxis: float}
+    """
+
+    data: dict[str, Any]


Maybe you can create a pydantic model per type and let pydantic figure out which one matches the parameters in data?

cfs-data · 2026-01-19T14:45:45Z

packages/scratch-core/src/conversion/data_formats.py


    scan_image: ScanImage
    mark_type: MarkType
+    crop_infos: list[CropInfo]


Can we use a tuple here instead of a list?

cfs-data · 2026-01-19T14:46:04Z

packages/scratch-core/src/conversion/mask.py

 def crop_to_mask(
    image: ScanMap2DArray,
    mask: MaskArray,
+    margin: Optional[int] = None,


Suggested change

margin: Optional[int] = None,

margin: int | None = None,

cfs-data · 2026-01-19T14:46:29Z

packages/scratch-core/src/conversion/mask.py

    Crops an image to the bounding box of a mask.

    :param image: The image to crop
    :param mask: Binary mask


Please add the new margin parameter to docstring

cfs-data · 2026-01-19T14:47:58Z

packages/scratch-core/src/conversion/mask.py


-def _determine_bounding_box(mask: MaskArray) -> tuple[slice, slice]:
+def _determine_bounding_box(
+    mask: MaskArray, margin: Optional[int] = None


Suggested change

mask: MaskArray, margin: Optional[int] = None

mask: MaskArray, margin: int | None = None

SimoneAriens · 2026-01-19T14:50:20Z

packages/scratch-core/src/conversion/rotate.py

+    :param times_median: Parameter used to determine what is considered an outlier when removing outliers/needles.
+    :return: Tuple of the cropped, rotated and masked scan image, and the rotated and cropped mask.
+    """
+    rotation_angle = get_rotation_angle(crop_infos, rotation_angle)


according to this function, crop_infos cannot be None

cfs-data · 2026-01-19T14:53:29Z

packages/scratch-core/src/conversion/remove_needles.py

+        )
+
+        # Apply mask and prepare data
+        scan_image_masked = ScanImage(


There is a mask_and_crop_scan_image method for this

cfs-data · 2026-01-19T14:53:47Z

packages/scratch-core/src/conversion/remove_needles.py

+    :param times_median: Parameter to help determine the outlier threshold.
+    :return: The cleaned scan image.
+    """
+    filter_size_moderated = 5


Make this a module constant, or parameter with default value?

cfs-data · 2026-01-19T14:54:11Z

packages/scratch-core/src/conversion/remove_needles.py

+    times_median = times_median * 6
+
+    # Check if this is a small strip of data
+    is_small_strip = scan_image.width <= 20 or scan_image.height <= 20


Change the magic value 20 into a module constant?

cfs-data · 2026-01-19T15:00:27Z

packages/scratch-core/src/conversion/mask.py


    :param mask: Binary mask array
+    :param margin: Margin around the bounding box to either crop (positive) or extend (negative) the bounding box
    :return: Tuple of (y_slice, x_slice) for the bounding box


Order of the tuple in return type here is not matching the output

SimoneAriens · 2026-01-19T15:02:58Z

packages/scratch-core/src/conversion/mask.py



-def _determine_bounding_box(mask: MaskArray) -> tuple[slice, slice]:
+def _determine_bounding_box(


Suggested change

def _determine_bounding_box(

def determine_bounding_box(

SimoneAriens · 2026-01-19T15:05:12Z

packages/scratch-core/src/conversion/mask.py


    :param mask: Binary mask array
+    :param margin: Margin around the bounding box to either crop (positive) or extend (negative) the bounding box
    :return: Tuple of (y_slice, x_slice) for the bounding box


switch as well

SimoneAriens · 2026-01-19T15:16:07Z

packages/scratch-core/src/conversion/mask.py

    y_max, x_max = np.max(non_zero_coords, axis=1)
-    return slice(x_min, x_max + 1), slice(y_min, y_max + 1)
+
+    if margin:


if margin defaults to 0 then this is cleaner:

y_coords, x_coords = np.nonzero(mask) y_min = max(0, y_coords.min() + margin) y_max = min(mask.shape[0], y_coords.max() - margin + 1) x_min = max(0, x_coords.min() + margin) x_max = min(mask.shape[1], x_coords.max() - margin + 1)

cfs-data · 2026-01-19T15:17:33Z

packages/scratch-core/src/conversion/remove_needles.py

+    scan_image: ScanImage, mask: MaskArray, times_median: float = 15.0
+) -> ScanImage:
+    """
+    Remove needle artifacts (outliers) from depth measurement data using median filtering.


What are needle artifacts?

cfs-data · 2026-01-19T15:23:19Z

packages/scratch-core/src/conversion/remove_needles.py

+        # Calculate subsampling factor for computational efficiency
+        # Goal: 7 μm sampling with 70 μm filter diameter
+        subsample_factor = int(
+            np.ceil(70e-6 / filter_size_moderated / scan_image.scale_x)


Magic value should be module constant

cfs-data · 2026-01-19T15:23:41Z

packages/scratch-core/src/conversion/remove_needles.py

+
+    if not is_small_strip:
+        # Calculate subsampling factor for computational efficiency
+        # Goal: 7 μm sampling with 70 μm filter diameter


"sampling" is not clear here, do you mean pixel size / scale?

cfs-data · 2026-01-19T15:24:41Z

packages/scratch-core/src/conversion/remove_needles.py

+        if subsample_factor > 1:
+            scan_image_subsampled = unwrap_result(
+                subsample_scan_image(
+                    scan_image_masked, subsample_factor, subsample_factor


Please use named keywords when there are multiple keywords

cfs-data · 2026-01-19T15:26:46Z

packages/scratch-core/src/conversion/remove_needles.py

+
+        residual_image = (
+            scan_image_masked.data
+            - scan_image_filtered.data[


Is scan_image_filtered containing NaNs in the same places as scan_image_masked?

cfs-data · 2026-01-19T15:27:38Z

packages/scratch-core/src/conversion/remove_needles.py

+            ]
+        )
+    else:
+        # For small strips: use 1D filtering with adjusted kernel size


This function is now very long. I think it would be beneficial to split the function up into smaller functions with single responsibilities, and only create a small pipeline / logical flow here

cfs-data · 2026-01-19T15:31:57Z

packages/scratch-core/src/conversion/remove_needles.py

+
+    # Find outliers: points where |residual| > threshold
+    threshold = times_median * np.nanmedian(np.abs(residual_image))
+    indices_invalid = np.abs(residual_image) > threshold


Suggested change

indices_invalid = np.abs(residual_image) > threshold

mask_outliers = np.abs(residual_image) > threshold

SimoneAriens · 2026-01-19T15:35:46Z

packages/scratch-core/src/conversion/remove_needles.py

+    times_median = times_median * 6
+
+    # Check if this is a small strip of data
+    is_small_strip = scan_image.width <= 20 or scan_image.height <= 20


here you are checking for both width and height (as opposed to the matlab code) but further down you are assuming the width is the shortest?

cfs-data · 2026-01-19T15:43:56Z

packages/scratch-core/src/conversion/remove_needles.py

+    )
+
+
+def apply_median_filter(scan_image: ScanImage, filter_size: int) -> ScanImage:


I think it's a good idea to rewrite this implementation in a way that makes more clear what actually happens. From the code it seems we are applying a windowing function and compute the medians over the windows?

cfs-data · 2026-01-19T15:45:30Z

packages/scratch-core/src/conversion/rotate.py

+def rotate_crop_and_mask_image_by_crop(
+    scan_image: ScanImage,
+    mask: MaskArray,
+    rotation_angle: float = 0.0,


Why does this function need a default value of 0.0?

SimoneAriens · 2026-01-19T15:56:52Z

packages/scratch-core/src/conversion/remove_needles.py

+    indices_invalid = np.abs(residual_image) > threshold
+
+    # Remove outliers by setting them to NaN
+    scan_image_without_outliers = scan_image.data.copy()


this is not a scan_image anymore. And I was wondering whether we could make a update_mark_data-like function to cover all this boilerplate up?

jokkalt and others added 30 commits January 6, 2026 12:37

ruff

bab246e

Add returns dependency

575d3e5

Returns will be use to wrap function in container that can be passed around in rail. Thus this will be use for railway function methodology

create railway logger decorator

73f41f8

Rewrite functions scratch-core with returns

3ba3c82

Writen module docstrings for new modules with the assistant of Claud (AI)

Create an abstract generic pipeline module

a9ec102

Pipelines is an abstracted module that communicates with the returns layer. This means endpoints can call scratch-core function without caring without caring what the underlying protocol is used for failures/exceptions

Create some preprocessors pipelines

c1bf7aa

Update preprocessors schemas

aeedebf

Create an upload scan parameters model that hold all the parameters to be used for the pipelines. All the parameters have a defaults and defaults will be generated if no parameters is given, thus nothing changes for the user or endpoint contract

Entegrate new pipeline structure to the pipeline

48d92c6

Cleanup old/unused scratch core code

315e0a3

Remove unused api code

ced309d

Make deptry happy

fba3bd8

deptry notify that dependencies needed maintainance

remove surface normals

6f44b47

mask and crop functionality added (#67)

df127e5

* mask and crop functionality added * type fixes and some efficiency

Gaussian filter - filter_apply.m (#60)

b5fe028

gaussian filter --------- Co-authored-by: Sharlon N. Regales <[email protected]>

Add map_level.m to translations (#66)

e03052f

Resampling method added (#71)

3d2a00f

* resampling method added

Feature/resampling mark (#72)

1252449

* Mark type and resampling for marks implemented

eod

cea1b14

ruff

73cffce

Some updates to resampling (#76)

95504b0

tests for rotation angle

bb829d6

rotation angle tests, fixtures and code

7cadb2f

added testfiles

871f4bc

added files to lfs

ruff

64d00d9

some change

e38dc16

Merge branch 'main' into feature/rotate_cropped_image

872a04e

lock file from main

e29b9dd

rotate image; roughly

66579ef

too many ruff errros

39478f6

last docstring changes

2d751ff

jokkalt changed the title ~~WIP Feature/rotate cropped image~~ Feature/rotate cropped image Jan 15, 2026

jokkalt marked this pull request as ready for review January 15, 2026 15:05

jokkalt added 2 commits January 15, 2026 16:08

Merge branch 'main' into feature/rotate_cropped_image

9e223d7

helper_function mess

1e753be

SimoneAriens reviewed Jan 19, 2026

View reviewed changes

cfs-data reviewed Jan 19, 2026

View reviewed changes

SimoneAriens reviewed Jan 19, 2026

View reviewed changes

cfs-data reviewed Jan 19, 2026

View reviewed changes

SimoneAriens reviewed Jan 19, 2026

View reviewed changes

cfs-data reviewed Jan 19, 2026

View reviewed changes

SimoneAriens reviewed Jan 19, 2026

View reviewed changes

cfs-data reviewed Jan 19, 2026

View reviewed changes

SimoneAriens reviewed Jan 19, 2026

View reviewed changes

	mask: MaskArray, margin: Optional[int] = None
	mask: MaskArray, margin: int \| None = None



		def _determine_bounding_box(mask: MaskArray) -> tuple[slice, slice]:
		def _determine_bounding_box(

	indices_invalid = np.abs(residual_image) > threshold
	mask_outliers = np.abs(residual_image) > threshold

		)


		def apply_median_filter(scan_image: ScanImage, filter_size: int) -> ScanImage:

Feature/rotate cropped image #82

Are you sure you want to change the base?

Feature/rotate cropped image #82

Uh oh!

Conversation

jokkalt commented Jan 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cfs-data Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cfs-data Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cfs-data Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

jokkalt commented Jan 8, 2026 •

edited

Loading

cfs-data Jan 19, 2026 •

edited

Loading

cfs-data Jan 19, 2026 •

edited

Loading

cfs-data Jan 19, 2026 •

edited

Loading