Change of dataset!

app.py

@@ -26,10 +26,9 @@ def preprocess_and_mosaic(image: Image.Image, grid_size: int, n_colors: int):
     
 
     cell_size = 400 // grid_size
-    actual_grid_size = 400 // cell_size  # This might be smaller than grid_size
-    actual_image_size = actual_grid_size * cell_size  # This might be smaller than 400
+    actual_grid_size = 400 // cell_size
+    actual_image_size = actual_grid_size * cell_size
     
-    # Crop the image to match the actual grid dimensions
     resized_np_cropped = resized_np[:actual_image_size, :actual_image_size]
     
     grid_labels, seg_time = timed(segment_image_grid)(resized_np_cropped, actual_grid_size, color_centers)
@@ -82,7 +81,7 @@ with gr.Blocks() as demo:
     with gr.Row():
         orig_out = gr.Image(label="Resized & Color Quantized (400x400)")
         seg_out = gr.Image(label="Grid Segmentation Preview")
-        mosaic_out = gr.Image(label="Flower Mosaic Result")
+        mosaic_out = gr.Image(label="Mosaic Result")
     info_out = gr.Textbox(label="Metrics & Timing", interactive=False)
     
     # Add examples using Unsplash URLs

createTiles.py

@@ -0,0 +1,95 @@
+import os
+import requests
+import io
+import colorsys
+import random
+from PIL import Image
+from datasets import load_dataset
+
+def rgb_to_hsv(rgb):
+    """Convert RGB [0-255] to HSV tuple (h, s, v)."""
+    r, g, b = [c / 255.0 for c in rgb]
+    return colorsys.rgb_to_hsv(r, g, b)  # values in [0,1]
+
+def dominant_hue(palettes):
+    """Return dominant hue (0–360) from palettes dict of RGB lists."""
+    all_colors = []
+    for k, colors in palettes.items():
+        all_colors.extend(colors)
+    if not all_colors:
+        return None
+    hsv_vals = [rgb_to_hsv(c) for c in all_colors]
+    hues = [h * 360 for (h, s, v) in hsv_vals if s > 0.2 and v > 0.2]  # ignore gray/dark
+    if not hues:
+        return None
+    return sum(hues) / len(hues)
+
+def create_balanced_tiles(limit=500, tile_size=64, bins=12):
+    """Create a balanced set of colored tiles from Unsplash-Lite-Palette."""
+    os.makedirs("data/tiles", exist_ok=True)
+
+    print("Loading Unsplash-Lite-Palette dataset...")
+    ds = load_dataset("1aurent/unsplash-lite-palette", split="train")
+
+    # Group images into hue bins
+    hue_bins = {i: [] for i in range(bins)}
+    for row in ds:
+        h = dominant_hue(row["palettes"])
+        if h is not None:
+            bin_idx = int(h / (360 / bins)) % bins
+            hue_bins[bin_idx].append(row)
+
+    print(f"Dataset grouped into {bins} hue bins.")
+
+    # Pick evenly from bins
+    per_bin = max(1, limit // bins)
+    selected = []
+    for b in range(bins):
+        random.shuffle(hue_bins[b])
+        selected.extend(hue_bins[b][:per_bin])
+
+    # In case fewer than limit (some bins are sparse)
+    if len(selected) < limit:
+        extra_needed = limit - len(selected)
+        all_rows = [row for bin_rows in hue_bins.values() for row in bin_rows]
+        random.shuffle(all_rows)
+        selected.extend(all_rows[:extra_needed])
+
+    print(f"Selected {len(selected)} images across hue bins.")
+
+    successful = 0
+    for i, row in enumerate(selected):
+        url = row["url"]  # direct image URL
+        filename = f"tile_{i:04d}.jpg"
+
+        try:
+            resp = requests.get(url, timeout=10)
+            resp.raise_for_status()
+            img = Image.open(io.BytesIO(resp.content)).convert("RGB")
+
+            # Crop to square
+            w, h = img.size
+            if w != h:
+                min_side = min(w, h)
+                left = (w - min_side) // 2
+                top = (h - min_side) // 2
+                img = img.crop((left, top, left + min_side, top + min_side))
+
+            # Resize
+            img = img.resize((tile_size, tile_size), Image.Resampling.LANCZOS)
+
+            # Save
+            filepath = os.path.join("data/tiles", filename)
+            img.save(filepath, "JPEG", quality=85, optimize=True)
+            successful += 1
+
+            if i % 50 == 0:
+                print(f"✓ {successful}/{i+1} saved ({filename})")
+
+        except Exception as e:
+            print(f"✗ Failed {i} ({url}): {e}")
+
+    print(f"\n✅ Done! {successful} balanced color tiles saved in data/tiles/")
+
+if __name__ == "__main__":
+    create_balanced_tiles(limit=500, tile_size=64, bins=12)