Upload agents/modality_detector.py with huggingface_hub

agents/modality_detector.py

@@ -121,9 +121,6 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     sharp_region = sharpness > sharp_thresh
     blur_region = sharpness < blur_thresh
     
-    # Compute transition sharpness
-    # In portrait mode: boundary between sharp/blur is very steep
-    # In real DoF: boundary is gradual
     sharp_fraction = float(np.mean(sharp_region))
     blur_fraction = float(np.mean(blur_region))
     
@@ -131,14 +128,38 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     blur_values = sharpness[blur_region] if np.any(blur_region) else np.array([0])
     blur_uniformity = 1.0 - min(float(np.std(blur_values)) / (float(np.mean(blur_values)) + 1e-9), 1.0)
     
+    # Check transition abruptness: in portrait mode the edge between sharp/blur is steep
+    # Compute gradient of sharpness map
+    sharpness_grad = np.hypot(
+        sobel(sharpness, axis=0),
+        sobel(sharpness, axis=1)
+    )
+    # High max gradient at the transition = abrupt = computational
+    max_transition = float(np.percentile(sharpness_grad, 99))
+    mean_transition = float(np.mean(sharpness_grad))
+    transition_abruptness = max_transition / (mean_transition + 1e-9)
+    
     indicators["sharp_fraction"] = round(sharp_fraction, 3)
     indicators["blur_fraction"] = round(blur_fraction, 3)
     indicators["blur_uniformity"] = round(blur_uniformity, 3)
-    
-    # Strong portrait mode signal: distinct sharp/blur regions with uniform blur
-    if blur_fraction > 0.3 and blur_uniformity > 0.6 and sharp_fraction > 0.15:
-        scores["PORTRAIT_MODE"] = scores.get("PORTRAIT_MODE", 0) + 0.5
+    indicators["transition_abruptness"] = round(transition_abruptness, 3)
+    
+    # Portrait mode detection — relaxed thresholds + multiple signals
+    portrait_signals = 0
+    if blur_fraction > 0.2 and sharp_fraction > 0.1:
+        portrait_signals += 1
+    if blur_uniformity > 0.5:
+        portrait_signals += 1
+    if transition_abruptness > 5:  # Abrupt edge = computational segmentation
+        portrait_signals += 1
+    
+    # Strong portrait mode: at least 2 of 3 signals
+    if portrait_signals >= 2:
+        scores["PORTRAIT_MODE"] = scores.get("PORTRAIT_MODE", 0) + 0.3 * portrait_signals
         indicators["portrait_mode_signature"] = True
+    elif portrait_signals == 1 and blur_fraction > 0.25:
+        scores["PORTRAIT_MODE"] = scores.get("PORTRAIT_MODE", 0) + 0.2
+        indicators["portrait_mode_weak"] = True
     
     # ── 4. Screenshot detection ───────────────────────────────────────
     # Screenshots have: perfect pixel edges, UI elements, uniform background areas
@@ -179,14 +200,29 @@ def detect_modality(img: Image.Image) -> ModalityResult:
         modality = max(scores, key=scores.get)
         confidence = min(1.0, scores[modality])
     
-    # Override: if portrait mode + smartphone, portrait mode wins
-    if scores.get("PORTRAIT_MODE", 0) > 0.3 and scores.get("SMARTPHONE", 0) > 0:
+    # Override: portrait mode wins over messaging/smartphone when detected
+    # (a portrait photo sent via WhatsApp should get PORTRAIT_MODE adjustments,
+    #  which are a superset of MESSAGING adjustments for the tests that matter)
+    if scores.get("PORTRAIT_MODE", 0) > 0.2:
         modality = "PORTRAIT_MODE"
-        confidence = min(1.0, scores["PORTRAIT_MODE"] + scores["SMARTPHONE"] * 0.3)
+        confidence = min(1.0, scores["PORTRAIT_MODE"])
     
-    # ── 7. Build score adjustments per modality ───────────────────────
+    # ── 7. Build score adjustments ────────────────────────────────────
+    # Merge adjustments when multiple modalities detected
+    # (e.g., portrait mode photo sent via messaging app gets BOTH sets)
     adjustments = _get_modality_adjustments(modality)
     
+    # If messaging signals are present alongside portrait mode, merge messaging adjustments
+    if modality == "PORTRAIT_MODE" and scores.get("MESSAGING", 0) > 0.2:
+        messaging_adj = _get_modality_adjustments("MESSAGING")
+        for test_name, multiplier in messaging_adj.items():
+            if test_name not in adjustments:
+                adjustments[test_name] = multiplier
+            else:
+                # Take the more suppressive (lower) multiplier
+                adjustments[test_name] = min(adjustments[test_name], multiplier)
+        indicators["dual_modality"] = "PORTRAIT_MODE + MESSAGING"
+    
     return ModalityResult(
         modality=modality,
         confidence=round(confidence, 3),