Update app.py

app.py

@@ -1,65 +1,169 @@
 # app.py
 import gradio as gr
 import cv2
+import numpy as np
+from PIL import Image, ExifTags
+import os
+
 from periodontitis_detection import SimpleDentalSegmentationNoEnhance
 
 # ==========================
-# 1️⃣ Load models once
+# Load model
 # ==========================
 model = SimpleDentalSegmentationNoEnhance(
-    unet_model_path="unet.keras",   # same filenames as your repo
+    unet_model_path="unet.keras",
     yolo_model_path="best2.pt"
 )
 
+# ====================================================
+# 1. Read DPI from metadata (EXIF / PNG)
+# ====================================================
+def read_dpi(path):
+    try:
+        img = Image.open(path)
+        info = img.info
+
+        # PIL standard DPI field
+        if "dpi" in info:
+            d = info["dpi"]
+            if isinstance(d, (tuple, list)):
+                return float(d[0])
+            return float(d)
+
+        # EXIF resolution (rare on xrays)
+        exif = img._getexif()
+        if exif:
+            for tag_id, value in exif.items():
+                tag = ExifTags.TAGS.get(tag_id, tag_id)
+                if tag == "XResolution":
+                    if isinstance(value, tuple) and value[1] != 0:
+                        return float(value[0]) / float(value[1])
+                    return float(value)
+    except:
+        pass
+
+    return None
+
+
+# ====================================================
+# 2. Detect 1 mm tick spacing
+# ====================================================
+def detect_tick_mm(path):
+    try:
+        img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)
+        if img is None:
+            return None
+
+        h, w = img.shape
+
+        # Right-side crop (where ruler usually is)
+        crop = img[:, int(w * 0.80):]
+
+        # Threshold for tick marks
+        blur = cv2.GaussianBlur(crop, (5, 5), 0)
+        _, thr = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
+
+        edges = cv2.Canny(thr, 50, 150)
+
+        lines = cv2.HoughLinesP(edges, 1, np.pi/180, threshold=30,
+                                minLineLength=10, maxLineGap=5)
+        if lines is None:
+            return None
+
+        ys = []
+        for l in lines:
+            x1, y1, x2, y2 = l[0]
+            if abs(y1 - y2) <= 3 and abs(x2 - x1) > 5:
+                ys.append(y1)
+
+        if len(ys) < 3:
+            return None
+
+        ys = sorted(ys)
+        diffs = np.diff(ys)
+
+        diffs = diffs[(diffs > 2) & (diffs < h // 2)]
+        if len(diffs) == 0:
+            return None
+
+        px_per_mm = float(np.mean(diffs))
+        return px_per_mm
+    except:
+        return None
+
+
+# ====================================================
+# 3. Compute mm per pixel
+# ====================================================
+def compute_mm_per_pixel(path):
+    # A) Metadata DPI
+    dpi = read_dpi(path)
+    if dpi and dpi > 1:
+        return (25.4 / dpi), "metadata"
+
+    # B) Tick marks (1 mm)
+    tick = detect_tick_mm(path)
+    if tick and tick > 0:
+        return (1.0 / tick), "tickmarks"
+
+    # C) Fallback 300 DPI
+    return (25.4 / 300.0), "fallback"
+
+
 # ==========================
-# 2️⃣ Define wrapper for Gradio
+# Wrapped function
 # ==========================
 def detect_periodontitis(image_np):
-    """
-    Gradio sends image as a NumPy RGB array.
-    We temporarily save it to a file path since analyze_image() needs a path.
-    """
+    # Save temporary image for model + mm scaling
     temp_path = "temp_input.jpg"
     cv2.imwrite(temp_path, cv2.cvtColor(image_np, cv2.COLOR_RGB2BGR))
 
-    # Run full pipeline
+    # Run periodontitis detection
     results = model.analyze_image(temp_path)
 
-    # Convert OpenCV BGR → RGB for Gradio display
+    # Convert combined BGR → RGB for display
     combined_rgb = cv2.cvtColor(results["combined"], cv2.COLOR_BGR2RGB)
 
-    # Optional: summarize measurements for text output
+    # Compute mm scaling
+    mm_per_px, method = compute_mm_per_pixel(temp_path)
+
+    # Summaries — CLEAN (no method labels)
     summaries = []
     for tooth in results["distance_analyses"]:
-        tooth_id = tooth["tooth_id"]
+        tid = tooth["tooth_id"]
         analysis = tooth["analysis"]
+
         if analysis:
-            mean_d = analysis["mean_distance"]
-            summaries.append(f"Tooth {tooth_id}: mean={mean_d:.2f}px")
+            px = analysis["mean_distance"]
+            mm = px * mm_per_px
+            summaries.append(f"Tooth {tid}: {mm:.2f} mm")
         else:
-            summaries.append(f"Tooth {tooth_id}: no valid CEJ–ABC measurement")
+            summaries.append(f"Tooth {tid}: no valid CEJ–ABC measurement")
+
+    summary_text = "\n".join(summaries)
 
-    summary_text = "\n".join(summaries) if summaries else "No detections found."
+    # Remove temp
+    try:
+        os.remove(temp_path)
+    except:
+        pass
 
     return combined_rgb, summary_text
 
 
 # ==========================
-# 3️⃣ Build Gradio Interface
+# Gradio Interface
 # ==========================
 demo = gr.Interface(
     fn=detect_periodontitis,
     inputs=gr.Image(type="numpy", label="Upload Dental X-Ray"),
     outputs=[
         gr.Image(label="Final Annotated Image (YOLO + CEJ–ABC)"),
-        gr.Textbox(label="Analysis Summary"),
+        gr.Textbox(label="Analysis Summary (mm)"),
     ],
-    title="🦷 Periodontitis Detection & Analysis",
-    description=(
-        "Automatically detects teeth (YOLOv8), segments CEJ/ABC (U-Net), "
-        "and measures CEJ–ABC distances per tooth to assess bone loss."
-    ),
+    title="🦷 Periodontitis Detection & Analysis (mm accurate)",
+    description="Outputs CEJ–ABC distances in millimeters."
 )
 
 if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)
+    demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)