Feature: Adaptive size selection based on local detail

[VoX]

Summary

• New GradientMap class computes Sobel gradient magnitude of the target image, downsampled to a 32x32 grid, normalized to [0,1]. High values = edges/detail, low values = smooth regions.
• Circle.randomize() now uses gradient-weighted size selection when USE_ADAPTIVE_SIZE is enabled: small circles near edges, large circles in smooth areas. Uses exp(-4 * |sizeNorm - (1 - gradient)|) weighting.
• Circle.mutateShape() scales position perturbation by gradient (fine-tune near edges, explore broadly in smooth areas) and uses gradient-biased size selection during mutation.
• Wired into Worker and Model — gradient map computed once from target image at initialization.
• USE_ADAPTIVE_SIZE feature flag in AppConstants (default true).

Test Results

Visual comparisons for 200 shapes on three test images are committed to test-results/proposal3/:

Image	Target	Uniform	Adaptive	Gradient Map	Diff
photo_detail
nature
edges

Test Plan

• Gradient map produces high values at edges, low/zero in smooth areas
• Checkerboard image produces high gradient across most cells
• Size selection near edges favors small circles vs smooth areas
• Adaptive sizing produces lower or equal final error vs uniform (within 5% tolerance)
• No regression across gradient, edges, nature, and photo_detail test images
• ./gradlew clean build passes

Implements Proposal 3 from PROPOSALS.md.

[Copilot]

Pull request overview

Adds “Proposal 3” adaptive circle sizing driven by a precomputed Sobel gradient map of the target image, aiming to use smaller circles near detailed/edge regions and larger circles in smooth areas.

Changes:

• Introduces GradientMap to compute a normalized, coarse-grid Sobel magnitude map and provide gradient-weighted size selection + mutation scaling.
• Wires adaptive sizing into Model initialization and Worker/Circle shape randomization & mutation under a new USE_ADAPTIVE_SIZE flag.
• Adds an end-to-end JUnit test suite and commits visual benchmark artifacts under test-results/proposal3/.

Reviewed changes

Copilot reviewed 7 out of 22 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
test-results/proposal3/photo_detail_uniform.png	Adds visual benchmark artifact for uniform sizing (photo_detail).
test-results/proposal3/photo_detail_target.png	Adds target image artifact for photo_detail benchmark.
test-results/proposal3/photo_detail_gradient.png	Adds gradient-map visualization artifact for photo_detail.
test-results/proposal3/photo_detail_diff.png	Adds diff heatmap artifact (uniform vs adaptive) for photo_detail.
test-results/proposal3/photo_detail_adaptive.png	Adds visual benchmark artifact for adaptive sizing (photo_detail).
test-results/proposal3/nature_uniform.png	Adds visual benchmark artifact for uniform sizing (nature).
test-results/proposal3/nature_target.png	Adds target image artifact for nature benchmark.
test-results/proposal3/nature_gradient.png	Adds gradient-map visualization artifact for nature.
test-results/proposal3/nature_diff.png	Adds diff heatmap artifact (uniform vs adaptive) for nature.
test-results/proposal3/nature_adaptive.png	Adds visual benchmark artifact for adaptive sizing (nature).
test-results/proposal3/edges_uniform.png	Adds visual benchmark artifact for uniform sizing (edges).
test-results/proposal3/edges_target.png	Adds target image artifact for edges benchmark.
test-results/proposal3/edges_gradient.png	Adds gradient-map visualization artifact for edges.
test-results/proposal3/edges_diff.png	Adds diff heatmap artifact (uniform vs adaptive) for edges.
test-results/proposal3/edges_adaptive.png	Adds visual benchmark artifact for adaptive sizing (edges).
src/test/java/com/bobrust/generator/AdaptiveSizeSelectionTest.java	Adds correctness + end-to-end regression tests and generates benchmark images.
src/main/resources/version	Bumps app version to 0.6.80.
src/main/java/com/bobrust/util/data/AppConstants.java	Adds USE_ADAPTIVE_SIZE feature flag (default true).
src/main/java/com/bobrust/generator/Worker.java	Stores and exposes a GradientMap for use during shape generation/mutation.
src/main/java/com/bobrust/generator/Model.java	Computes gradient map once at initialization and wires it into Worker.
src/main/java/com/bobrust/generator/GradientMap.java	New Sobel-based gradient map with gradient-biased size selection and mutation scaling.
src/main/java/com/bobrust/generator/Circle.java	Uses gradient map to bias size selection and scale position mutation near edges.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

selectSizeIndex allocates a new float[] weights on every call. This method is called frequently (e.g., from Circle.randomize() / mutateShape()), so the per-call allocation will create substantial GC pressure and slow down generation. Consider rewriting selection to avoid allocations (e.g., two-pass compute total then select without storing the array, or reuse a preallocated array via ThreadLocal).

Suggested change

	float[] weights = newfloat[numSizes];
	floattotalWeight = 0;
	for (inti = 0; i < numSizes; i++) {
	floatsizeNorm = (float) i / (numSizes - 1); // 0=smallest, 1=largest
	// High gradient -> prefer small (low sizeNorm), low gradient -> prefer large
	weights[i] = (float) Math.exp(-4.0 * Math.abs(sizeNorm - (1.0 - gradient)));
	totalWeight += weights[i];
	}
	// Weighted random selection
	floatr = rnd.nextFloat() * totalWeight;
	floatcumulative= 0;
	for (inti = 0;i< numSizes; i++) {
	cumulative += weights[i];
	floattotalWeight = 0;
	floattargetSizeNorm = 1.0f - gradient;
	for (inti = 0; i < numSizes; i++) {
	floatsizeNorm = (float) i / (numSizes - 1); // 0=smallest, 1=largest
	// High gradient -> prefer small (low sizeNorm), low gradient -> prefer large
	totalWeight += (float) Math.exp(-4.0 * Math.abs(sizeNorm - targetSizeNorm));
	}
	// Weighted random selection without per-call array allocation
	floatr = rnd.nextFloat() * totalWeight;
	floatcumulative = 0;
	for (inti= 0;i < numSizes; i++) {
	floatsizeNorm = (float)i/ (numSizes - 1); // 0=smallest, 1=largest
	cumulative += (float) Math.exp(-4.0 * Math.abs(sizeNorm - targetSizeNorm));

[Copilot]

This test writes generated images into test-results/proposal3, which is not ignored by git and is also committed in this PR. Running the test will overwrite tracked files and/or leave a dirty working tree. Consider writing outputs under build/ (as ErrorGuidedPlacementTest does) or gating the visual-output portion behind a system property / @Disabled so regular test runs don’t mutate the repository.

[Copilot]

Unused import: java.util.Arrays is imported but not used in this test class. Removing it will keep the test clean and avoid IDE/compiler warnings.

Suggested change

import java.util.Arrays;