UPDATE: DOMINANT COLOR

app.py

@@ -49,11 +49,66 @@ def depth_to_normal(depth):
     normal = (normal + 1) / 2
     return normal
 
+# ===============================
+# NEW: Dominant color from non-transparent pixels
+# ===============================
+def dominant_color_from_image(img_rgba: Image.Image, max_samples: int = 10000, k: int = 3):
+    """
+    Returns (rgb_tuple, hex_string) for the dominant color considering only alpha>0 pixels.
+    Uses K-means on a random subset of pixels for speed.
+    """
+    arr = np.array(img_rgba.convert("RGBA"))
+    alpha = arr[..., 3]
+    mask = alpha > 0
+
+    if not np.any(mask):
+        # Fallback if everything is transparent
+        rgb = (255, 255, 255)
+        hex_color = "#{:02X}{:02X}{:02X}".format(*rgb)
+        return rgb, hex_color
+
+    rgb_pixels = arr[mask][:, :3].astype(np.float32)
+
+    # Subsample for speed
+    if rgb_pixels.shape[0] > max_samples:
+        idx = np.random.choice(rgb_pixels.shape[0], max_samples, replace=False)
+        rgb_pixels = rgb_pixels[idx]
+
+    # K-means
+    criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 20, 1.0)
+    attempts = 3
+    compactness, labels, centers = cv2.kmeans(
+        data=rgb_pixels,
+        K=k,
+        bestLabels=None,
+        criteria=criteria,
+        attempts=attempts,
+        flags=cv2.KMEANS_PP_CENTERS
+    )
+
+    # Pick the largest cluster
+    counts = np.bincount(labels.flatten(), minlength=k)
+    dominant_idx = int(np.argmax(counts))
+    dom = centers[dominant_idx].clip(0, 255).astype(np.uint8)
+    rgb = (int(dom[0]), int(dom[1]), int(dom[2]))
+    hex_color = "#{:02X}{:02X}{:02X}".format(*rgb)
+    return rgb, hex_color
+
 # ===============================
 # CORE PROCESSING FUNCTION
 # ===============================
 def _process_saree_core(base_image: Image.Image, pattern_image: Image.Image):
-    img_pil = base_image.convert("RGB")
+    # ORIGINAL (kept as comment): This converts alpha to black (0,0,0)
+    # img_pil = base_image.convert("RGB")
+
+    # NEW: Matte transparent pixels to the dominant color from the image itself
+    base_rgba = base_image.convert("RGBA")
+    fill_rgb, fill_hex = dominant_color_from_image(base_rgba)
+    bg = Image.new("RGBA", base_rgba.size, fill_rgb + (255,))
+    img_pil = Image.alpha_composite(bg, base_rgba).convert("RGB")  # <- alpha-aware RGB conversion
+    # (Optional) If you want to inspect which color was used:
+    # print("Dominant matte color:", fill_hex)
+
     img_np = np.array(img_pil)
 
     # Prepare tensor
@@ -129,7 +184,6 @@ def _process_saree_core(base_image: Image.Image, pattern_image: Image.Image):
     mask_binary = (mask_alpha > k/100).astype(np.uint8) * 255  # slightly stricter threshold
     mask_eroded = cv2.erode(mask_binary, kernel, iterations=3)  # balanced erosion
 
-
     # 2. Feather edges (blur)
     mask_blurred = cv2.GaussianBlur(mask_eroded, (15, 15), sigmaX=3, sigmaY=3)
 
@@ -268,4 +322,4 @@ async def unhandled_exception_handler(request: Request, exc: Exception):
 
 if __name__ == "__main__":
     import uvicorn
-    uvicorn.run(app, host="0.0.0.0", port=7860)
+    uvicorn.run(app, host="0.0.0.0", port=7860)