Upload agents/modality_detector.py with huggingface_hub

agents/modality_detector.py

@@ -30,6 +30,25 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     
     w, h = img.size
     gray = np.array(img.convert("L")).astype(np.float64)
+    rgb = np.array(img.convert("RGB")).astype(np.float64)
+    
+    # ═══ CRITICAL PRE-CHECK: Bayer CFA pattern ═══════════════════════
+    # Real camera images have Bayer demosaicing traces: σ_green < σ_red ≈ σ_blue
+    # If this is absent, the image CANNOT be from a real camera sensor.
+    # This blocks portrait mode suppression from protecting AI images.
+    noise_std = {}
+    for c, nm in enumerate(["red", "green", "blue"]):
+        ch = rgb[:, :, c]
+        dn = gaussian_filter(ch, sigma=1.5)
+        noise_std[nm] = float(np.std(ch - dn))
+    
+    has_bayer = noise_std["green"] < min(noise_std["red"], noise_std["blue"])
+    bayer_margin = min(noise_std["red"], noise_std["blue"]) - noise_std["green"]
+    indicators["has_bayer"] = has_bayer
+    indicators["bayer_margin"] = round(bayer_margin, 4)
+    indicators["noise_g"] = round(noise_std["green"], 3)
+    indicators["noise_r"] = round(noise_std["red"], 3)
+    indicators["noise_b"] = round(noise_std["blue"], 3)
     
     # ═══ CONTENT-BASED DETECTION (works without metadata) ═════════════
     
@@ -75,18 +94,22 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     indicators["transition_abruptness"] = round(transition, 2)
     indicators["has_detail"] = has_detail
     
-    # Portrait mode scoring — multiple independent signals
+    # Portrait mode scoring — requires BOTH content signals AND real camera evidence
+    # CRITICAL: If no Bayer pattern, this CANNOT be a smartphone portrait photo.
+    # AI images that happen to have blurred backgrounds must NOT get portrait suppression.
+    can_be_portrait = has_detail and has_bayer
+    
     portrait_score = 0.0
-    if has_detail and bimodal_ratio > 1.0:
-        portrait_score += 0.25  # Strong bimodal sharpness
-    if has_detail and blur_uniformity > 0.5:
-        portrait_score += 0.2   # Uniform blur region
-    if has_detail and transition > 4.0:
-        portrait_score += 0.2   # Abrupt sharp/blur boundary
-    if has_detail and blur_frac > 0.2 and sharp_frac > 0.1:
-        portrait_score += 0.15  # Distinct regions exist
-    
-    if portrait_score > 0.3:
+    if can_be_portrait and bimodal_ratio > 1.0:
+        portrait_score += 0.25
+    if can_be_portrait and blur_uniformity > 0.5:
+        portrait_score += 0.2
+    if can_be_portrait and transition > 4.0:
+        portrait_score += 0.2
+    if can_be_portrait and blur_frac > 0.2 and sharp_frac > 0.1:
+        portrait_score += 0.15
+    
+    if portrait_score > 0.3 and has_bayer:
         scores["PORTRAIT_MODE"] = portrait_score
         indicators["portrait_detected"] = True
     
@@ -174,12 +197,20 @@ def detect_modality(img: Image.Image) -> ModalityResult:
         modality = "PORTRAIT_MODE"
         conf = min(1.0, scores["PORTRAIT_MODE"])
     
-    # SAFETY GUARD: No detail = possible AI. Disable all suppression.
+    # SAFETY GUARD 1: No detail = possible AI. Disable all suppression.
     if not has_detail:
         modality = "UNKNOWN"
         conf = 0.2
         indicators["safety_override"] = "Low-detail image — suppression disabled"
     
+    # SAFETY GUARD 2: No Bayer pattern = not from a real camera sensor.
+    # AI-generated images lack Bayer demosaicing traces.
+    # Never apply camera-related suppression without Bayer evidence.
+    if not has_bayer and modality in ("PORTRAIT_MODE", "SMARTPHONE", "MESSAGING"):
+        modality = "UNKNOWN"
+        conf = 0.2
+        indicators["safety_override"] = f"No Bayer CFA pattern (margin={bayer_margin:.3f}) — not from a real camera sensor. All suppression disabled."
+    
     # ═══ BUILD ADJUSTMENTS ════════════════════════════════════════════
     
     adjustments = _get_modality_adjustments(modality)