fix Overlap to require expandOverlap parameter (TODO: remove all old … - #106
Open
StephanPreibisch wants to merge 4 commits into
Open
fix Overlap to require expandOverlap parameter (TODO: remove all old …#106StephanPreibisch wants to merge 4 commits into
StephanPreibisch wants to merge 4 commits into
Conversation
…fusion code in the transformed package - unrelated) more importantly, fixing issues reported in #104 in a different way, but providing a guess to the inverse TPS. For this we made a compound transformation model that first applies the guess, and then runs the InverseRealTransformGradientDescent, which takes a guess in inverseTol() provided by the source[] in apply().
…ting view bounding boxes imglib2's InverseRealTransformGradientDescent.apply(s, t) calls inverseTol(s, s, ...), i.e. it seeds the descent with the query point itself. For a tile whose render position is tens of thousands of pixels from its local pixel coordinates, and with only a few landmarks, the descent never reaches the true preimage and the resulting bounding box silently drops the view from every block it should contribute to. GuessingRealTransform now calls inverseTol(p, invGuess(p), ...) directly for every corner, using the per-view approximate affine (image -> render) as the seed. Composing the guess with WrappedIterativeInvertibleRealTransform.inverse() cannot work: that ends in inverseTol(guess, guess, ...) and inverts a different point. A corner is trusted only if the descent converged and stayed within max(dimensions) of the guess; otherwise the guess is used and reported. init() now passes the approximate affine itself (not its inverse) as the guess. fitAffineTransform throws instead of silently returning an identity when the landmarks are ill-defined (e.g. co-planar). Adds InverseTransformedBoundingBoxTest: exact ground truth with dense landmarks, sparse split-center landmarks where the un-seeded inverse is ~94,000 px off, and aliasing / copy() behaviour of GuessingRealTransform. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ate affine and bounding box The local init() and the BigStitcher-Spark displacement-field cache both need the same three things per underlying view: the TPS (render -> image), the affine approximating its inverse (image -> render) and the back-projected bounding box seeded with that affine. Having each caller assemble them by hand is how the transform directions got mixed up. TpsSetup / setupTps() now does it once; init() uses it and reuses the TPS instance for displacement-field sampling instead of solving it a second time. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
| IntervalSamplingMethod.CORNERS ) ); | ||
|
|
||
| if ( invTransform.numFallbacks() > 0 ) | ||
| IOFunctions.println( "[TPS] inverseTransformedBoundingBox: " + invTransform.numFallbacks() + " of " |
Collaborator
There was a problem hiding this comment.
Just an FYI: IOFunctions.println breaks on headless spark (I might use System.out.println instead)
indecisiveuser
approved these changes
Sep 3, 2026
Collaborator
|
Looks good to me! Byte identical to #104 on my real-data test. |
Closed
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
2 participants
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.
…fusion code in the transformed package - unrelated)
more importantly, fixing issues reported in #104 in a different way, but providing a guess to the inverse TPS. For this we made a compound transformation model that first applies the guess, and then runs the InverseRealTransformGradientDescent, which takes a guess in inverseTol() provided by the source[] in apply().