Add style features, CLIP comparison, and experiment analysis#52
Merged
exdysa merged 2 commits intodarkshapes:mainfrom Mar 23, 2026
Merged
Add style features, CLIP comparison, and experiment analysis#52exdysa merged 2 commits intodarkshapes:mainfrom
exdysa merged 2 commits intodarkshapes:mainfrom
Conversation
New feature_style.py extracts 15 art-specific features (stroke patterns, color palette, composition, micro-texture) for AI artwork detection. EXPERIMENTS.md documents all feature experiments run on the Hemg AI-Art vs Real-Art dataset (4000 images, 5-fold CV), with results: Artwork features (49): 79.4% acc, 0.886 AUC Style features (15): 78.8% acc, 0.883 AUC Art + Style combined (64): 83.5% acc, 0.923 AUC CLIP ViT-B/32 (512): 89.3% acc, 0.963 AUC All combined (576): 90.0% acc, 0.966 AUC Key finding: CLIP embeddings outperform all hand-crafted features by +10pp. Combining hand-crafted features with CLIP adds only +0.7pp. See results/EXPERIMENTS.md for full analysis, code links, and limitations.
Reproducible test scripts for evaluating artwork detection features: - test_experiments.py: runs all 5 feature experiments (artwork, style, CLIP, combined) with 5-fold CV on Hemg dataset - test_fair_evaluation.py: tests on semantically-similar datasets to control for content confounds - test_scale_evaluation.py: measures accuracy vs training set size (400 to 4000 samples) - generate_fair_eval_pdf.py: generates timestamped PDF reports Run with: uv run python tests/test_experiments.py Includes result PDFs and JSON for reference. See EXPERIMENTS.md for interpretation.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
feature_style.py) — 15 features targeting brush strokes, color palettes, composition, and micro-textureuv run python tests/test_experiments.pyResults
Key finding: CLIP embeddings outperform all hand-crafted features by +10pp on fair art-vs-art data. See
results/EXPERIMENTS.mdfor full analysis, code pointers, limitations, and recommendations.Test plan
uv run pytest -vpassesuv run python tests/test_experiments.pyreproduces the results table above