Add rotomap automark3 command with DINOv3 support #529
Conversation
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
Implements issue #496 by adding a new 'rotomap automark3' command that uses DINOv3 for feature-based mole matching across images. Features: - New mel/lib/dinov3.py module for DINOv3 model loading and feature extraction - New mel/cmd/rotomapautomark3.py command with --reference and --target arguments - Supports multiple model sizes (small, base, large, giant) - Configurable cosine similarity threshold for match acceptance - Best-match aggregation when moles appear in multiple reference images - Updates non-canonical moles, skips canonical ones - Optional debug image output - Smoke test integration The command leverages DINOv3's improved semantic features for more accurate mole matching compared to previous versions.
Enhancements to automark3: - Add geometric position guessing when no existing mole is found - Uses mel.rotomap.relate.guess_mole_pos for initial position estimation - Combines geometric guessing (like guess-missing) with DINOv3 refinement - Improves ability to find completely missing moles New benchmark test: - test_benchmark_automark3: Tests automark3 on mel-datasets benchmark - Uses same dataset and metrics as guess-missing/guess-refine test - Expects to find all 3 removed moles and match at least 2 within 50 pixels - Validates automark3 performs at least as well as existing commands This makes automark3 a complete solution that can both find missing moles and refine their positions using DINOv3 features, without requiring a separate guess-missing step.
Fixes for new code (automark3 and dinov3): - Remove unused numpy import - Prefix unused variables with underscore (_feature_dim, _ref_x, etc.) - Remove unused torch import in search_for_mole function - Remove unnecessary old_x, old_y variables - Use ternary operator for feature extraction - Return directly instead of assigning to similarity variable Fixes for pre-existing code: - rotomapmergeextrastem.py: Prefix unused added_uuids with underscore - identifynn.py: Prefix unused i variables in training/validation steps - relate.py: Prefix unused to_points and error variables All code now passes ruff static analysis checks.
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.
Implements issue #496 by adding a new 'rotomap automark3' command that uses
DINOv3 for feature-based mole matching across images.
Features:
The command leverages DINOv3's improved semantic features for more accurate
mole matching compared to previous versions.