debug-palette.py

2026-05-11 17:08:27 +08:00 · 2026-02-02 19:35:11 -05:00
parent 58a91b312e
commit 979fda471a
1 changed files with 106 additions and 0 deletions
@@ -0,0 +1,106 @@
+#!/usr/bin/env python3
+"""
+Debug script to trace color extraction divergence between machines.
+Run this on both machines with the same image and compare outputs.
+
+Usage: python3 debug-palette.py /path/to/image.png
+"""
+
+import hashlib
+import json
+import sys
+from pathlib import Path
+
+from lib import read_image, ImageReadError
+from lib.color import rgb_to_lab
+from lib.palette import downsample_pixels, kmeans_cluster, _score_colors_count, _hue_to_family
+from lib.hct import Hct
+from lib.color import Color
+
+def hash_data(data: list) -> str:
+    """Create a short hash of list data for comparison."""
+    return hashlib.sha256(json.dumps(data, sort_keys=True).encode()).hexdigest()[:16]
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: python3 debug-palette.py /path/to/image.png")
+        sys.exit(1)
+
+    image_path = Path(sys.argv[1])
+
+    print(f"=== Debug Palette Extraction ===")
+    print(f"Image: {image_path}")
+    print(f"Python: {sys.version}")
+    print()
+
+    # Step 1: Read image
+    try:
+        pixels = read_image(image_path)
+    except ImageReadError as e:
+        print(f"Error reading image: {e}")
+        sys.exit(1)
+
+    print(f"1. Raw pixels: {len(pixels)} pixels")
+    print(f"   First 5: {pixels[:5]}")
+    print(f"   Hash: {hash_data(pixels[:1000])}")  # Hash first 1000 for speed
+    print()
+
+    # Step 2: Downsample
+    sampled = downsample_pixels(pixels, factor=4)
+    print(f"2. Downsampled: {len(sampled)} pixels")
+    print(f"   First 5: {sampled[:5]}")
+    print(f"   Hash: {hash_data(sampled)}")
+    print()
+
+    # Step 3: Convert to Lab
+    colors_lab = [rgb_to_lab(*c) for c in sampled]
+    print(f"3. Lab conversion:")
+    print(f"   First 5: {[tuple(round(v, 6) for v in lab) for lab in colors_lab[:5]]}")
+    # Round for hash to avoid float repr issues
+    lab_rounded = [[round(v, 10) for v in lab] for lab in colors_lab]
+    print(f"   Hash: {hash_data(lab_rounded)}")
+    print()
+
+    # Step 4: Sort by lightness (this is what k-means init does)
+    sorted_indices = sorted(range(len(colors_lab)), key=lambda i: colors_lab[i][0])
+    print(f"4. Sorted indices (by L):")
+    print(f"   First 10: {sorted_indices[:10]}")
+    print(f"   Hash: {hash_data(sorted_indices)}")
+    print()
+
+    # Step 5: K-means clustering (48 clusters for "count" mode)
+    clusters = kmeans_cluster(sampled, k=48)
+    print(f"5. K-means clusters: {len(clusters)} clusters")
+    print(f"   Top 5 by size:")
+    for i, (centroid, representative, count) in enumerate(clusters[:5]):
+        print(f"     {i+1}. centroid={centroid}, repr={representative}, count={count}")
+
+    # Hash cluster data
+    cluster_data = [(list(c), list(r), n) for c, r, n in clusters]
+    print(f"   Hash: {hash_data(cluster_data)}")
+    print()
+
+    # Step 6: Score colors (count mode = faithful)
+    colors_for_scoring = [(c[1], c[2]) for c in clusters]  # (representative, count)
+    scored = _score_colors_count(colors_for_scoring)
+    print(f"6. Scored colors (count/faithful mode):")
+    for i, (color, score) in enumerate(scored[:5]):
+        hct = color.to_hct()
+        family = _hue_to_family(hct.hue)
+        family_names = ["RED", "ORANGE", "YELLOW", "GREEN", "BLUE", "PURPLE"]
+        print(f"     {i+1}. {color.to_hex()} score={score:.0f} hue={hct.hue:.1f} chroma={hct.chroma:.1f} family={family_names[family]}")
+    print()
+
+    # Final result
+    print(f"=== Final Primary Color ===")
+    if scored:
+        primary = scored[0][0]
+        print(f"Primary: {primary.to_hex()}")
+        hct = primary.to_hct()
+        print(f"HCT: hue={hct.hue:.2f}, chroma={hct.chroma:.2f}, tone={hct.tone:.2f}")
+
+    print()
+    print("Compare the hashes between machines to find where divergence starts.")
+
+if __name__ == "__main__":
+    main()