Conversation
Plan: .github/plans/hough-harris-feature-detection-1-foundation.md Pass: initial implementation
Plan: .github/plans/hough-harris-feature-detection-1-foundation.md Pass: review follow-up
Plan: .github/plans/hough-harris-feature-detection-1-foundation.md Pass: metadata-only review follow-up
Plan: .github/plans/hough-harris-feature-detection-1-foundation.md Pass: review follow-up for RVW-002
Plan: .github/plans/hough-harris-feature-detection-1-foundation.md Pass: plan-bookkeeping closeout
Route standard line detection through the theta/rho accumulator while preserving the legacy SimpleHough.detectLines(...) compatibility surface.\n\nPlan: .github/plans/hough-harris-feature-detection-2-standard-hough-lines.md\nPass: initial implementation
Plan: .github/plans/hough-harris-feature-detection-2-standard-hough-lines.md Pass: plan-bookkeeping closeout
Adds Hough.detectLineSegments(...), the Vision.houghLineSegmentDetection(...) wrapper, and focused probabilistic fixtures/tests. Plan: .github/plans/hough-harris-feature-detection-3-probabilistic-hough-segments.md Pass: initial implementation
Restrict probabilistic duplicate merges to colinear fragments with small along-line gaps, correct merge construction on a shared axis, and add the adjacent-parallel regression requested in review. Plan: .github/plans/hough-harris-feature-detection-3-probabilistic-hough-segments.md Pass: review follow-up Finding: RVW-003
Reject mismatched custom edge-image inputs for probabilistic Hough segments, document the same-size wrapper contract, and add the focused wrapper regression requested in review. Plan: .github/plans/hough-harris-feature-detection-3-probabilistic-hough-segments.md Pass: review follow-up Finding: RVW-004
Record the approved step-3 outcome across the durable packet set, mark the step completed in the plan files, and activate step 4 without absorbing the unrelated Iterate.agent user edit. Plan: .github/plans/hough-harris-feature-detection-3-probabilistic-hough-segments.md Pass: plan-bookkeeping closeout
Add focused standard-Hough regressions for weighted votes, theta-window rejection, and detectLinesFromPoints parity, and record the explicit decision to defer multi-scale srn/stn parity from this step. Plan: .github/plans/hough-harris-feature-detection-4-hough-api-parity.md Pass: initial implementation
Finalize the approved step-4 bookkeeping by recording the normalized approval outcome, activating step 5 in the durable iteration state, and marking the plan set accordingly. Plan: .github/plans/hough-harris-feature-detection-4-hough-api-parity.md Pass: plan-bookkeeping closeout
Introduce the step-5 Hough circle implementation by adding the dedicated detector companion, routing the Hough and Vision circle APIs through it, and covering the new behavior with synthetic fixtures plus focused circle tests. Plan: .github/plans/hough-harris-feature-detection-5-hough-circles.md Pass: initial implementation
Resolve RVW-005 and RVW-006 by scaling Hough circle perimeter sampling with radius, removing the grayscale-as-edge fallback, and adding focused regressions for large-radius and no-edge inputs. Plan: .github/plans/hough-harris-feature-detection-5-hough-circles.md Pass: review follow-up
Record the approved step-5 Hough-circle outcome in the durable packet and progress state, mark the step-5 plan completed, and activate step 6 as the next implementation scope while preserving the unrelated Iterate.agent edit outside the commit. Plan: .github/plans/hough-harris-feature-detection-5-hough-circles.md Pass: plan-bookkeeping closeout
Implement the step-6 Harris response path with grayscale intensity sampling, separable derivative kernels, box or Gaussian local-tensor aggregation, and focused score-ordering regressions that keep synthetic corners above edge interiors and flat regions. Plan: .github/plans/hough-harris-feature-detection-6-harris-response.md Pass: initial implementation
Record the approved step-6 Harris response review in the durable packet, mark the step plan completed, activate step 7 in the ledger and progress note, and preserve the unrelated Iterate.agent user edit outside the commit scope. Plan: .github/plans/hough-harris-feature-detection-6-harris-response.md Pass: plan-bookkeeping closeout
Implement deterministic Harris corner extraction on top of the reviewed response map, expose the public Vision Harris wrappers, add focused corner-selection regressions, record HH-DEC-008, and preserve the unrelated Iterate.agent user edit plus the excluded orchestrator-owned run-ledger update outside the commit scope. Plan: .github/plans/hough-harris-feature-detection-7-harris-corners-and-api.md Pass: initial implementation
Record the approved step-7 review normalization, activate step 8 in the durable packet and progress state, update the plan set, and preserve the unrelated Iterate.agent user edit outside the commit scope. Plan: .github/plans/hough-harris-feature-detection-7-harris-corners-and-api.md Pass: plan-bookkeeping closeout
Refresh the public Vision docs and demo surface, add direct Harris facade coverage, sync the retained GeneratedSuites compatibility registry, and update the manual inventory for the final Hough/Harris closeout while keeping the unrelated Iterate.agent edit and orchestrator-owned run-ledger update out of scope. Plan: .github/plans/hough-harris-feature-detection-8-docs-tests-and-closeout.md Pass: initial implementation
Archive the completed Hough/Harris overview and subplans under .github/realized/, refresh the final queue-exhausted packet, progress, and execution-report state, and keep the unrelated Iterate.agent user edit out of scope. Plan: .github/realized/hough-harris-feature-detection-8-docs-tests-and-closeout.md Pass: queue-exhausted closeout
Plan: .github/plans/manual-review-integration.md Pass: initial implementation
Plan: .github/plans/manual-review-integration.md Pass: review follow-up
Plan: .github/plans/manual-review-integration.md Pass: review follow-up Record 89548ea as the latest committed bookkeeping anchor and adopt MRI-DEC-003 for the current metadata-only response.
Plan: .github/plans/manual-review-integration.md Pass: review follow-up Clarify the iteration workflow contract so the durable packet may stay anchored on 81fd569 while the current response is described separately.
Introduce new markdown templates for commit packets, decision logs, execution reports, implementation handoffs, review packets, run ledgers, and timelines to standardize documentation and facilitate smoother iteration and review workflows. Additionally, create placeholder files for agent progress and plans to enhance project organization.
Archive the completed Hough/Harris overview and all subplans under .github/realized/, update the final queue-exhausted packet, and ensure the unrelated Iterate.agent user edit remains out of scope. Plan: .github/realized/hough-harris-feature-detection-8-docs-tests-and-closeout.md Pass: queue-exhausted closeout
Merge split Hough/Harris modules into single algorithm files, return Ray2D lines and IntPoint2D corners, move visualization out of Hough, add HoughEdgeImageSizeMismatch, document private helpers, and refresh VisionMain demos plus unit tests. Co-authored-by: Cursor <cursoragent@cursor.com>
Plan: .cursor/plans/ci-portability-recovery-1-thread-fallback.md
Plan: .cursor/plans/ci-portability-recovery-2-portable-format-assertions.md
There was a problem hiding this comment.
doccomments are too technical, should follow other parts of the library where explanations ar eaimed at the average joe
|
|
||
| 1. **Response map** — convert the image to gradients, smooth structure-tensor terms | ||
| inside a local window, then score every pixel with | ||
| `det(M) - k * trace(M)^2`, where `M` is the summed gradient covariance. |
There was a problem hiding this comment.
Use pretty unicode representations for calculations like this using either Latex/pre md element (see Matrix2D/TransformationMatrix2D doc)
| /** | ||
| Allocates a zero-filled Harris response map with the given dimensions. | ||
| **/ | ||
| public static function createResponseMap(width:Int, height:Int):Matrix2D { |
There was a problem hiding this comment.
Seems pretty useless, matrix should probably have either a static method for creating a 0 filled matrix, or just have it a a constructor parameter of "fill". why bloat?
| binomial kernels, then windowed to produce `Ix^2`, `Iy^2`, and `Ix*Iy` sums. | ||
| Each pixel score is `det - k * trace^2`. | ||
| **/ | ||
| public static function computeResponse(image:Image, ?options:HarrisResponseOptions):Matrix2D { |
There was a problem hiding this comment.
This function is a sequence of steps, the doc doesnt actually say anything here to the average joe. the function is steps? the doc should be a step list as well.
| } | ||
|
|
||
| /** Clamps structure-tensor window size to at least 1. **/ | ||
| static inline function normalizeBlockSize(blockSize:Int):Int { |
There was a problem hiding this comment.
MathTools.clamp?/MathTools.min?
| for (y in 0...image.height) { | ||
| for (x in 0...image.width) { | ||
| var pixel = image.getPixel(x, y); | ||
| intensity.set(x, y, 0.2126 * pixel.red + 0.7152 * pixel.green + 0.0722 * pixel.blue); |
There was a problem hiding this comment.
What is this? why magic values? looks like grayscale magic to me, which already exists iirc. justify this.
Two major feature detection algorithms i wanted to implement for a looooooong time.
This still needs validation.