Update predictor.py

predictor.py

@@ -1,22 +1,43 @@
 import threading
 import numpy
+import opennsfw2
 from PIL import Image
+from keras import Model
 
 from roop.typing import Frame
 
-# Define any other necessary variables or constants here
+PREDICTOR = None
+THREAD_LOCK = threading.Lock()
+MAX_PROBABILITY = 0.85
+
+
+def get_predictor() -> Model:
+    global PREDICTOR
+
+    with THREAD_LOCK:
+        if PREDICTOR is None:
+            PREDICTOR = opennsfw2.make_open_nsfw_model()
+    return PREDICTOR
+
+
+def clear_predictor() -> None:
+    global PREDICTOR
+
+    PREDICTOR = None
+
 
 def predict_frame(target_frame: Frame) -> bool:
-    # Modify this function as needed for your specific use case, without NSFW prediction
-    # For example, you can implement custom image analysis or processing here
-    return False
+    image = Image.fromarray(target_frame)
+    image = opennsfw2.preprocess_image(image, opennsfw2.Preprocessing.YAHOO)
+    views = numpy.expand_dims(image, axis=0)
+    _, probability = get_predictor().predict(views)[0]
+    return probability > MAX_PROBABILITY
+
 
 def predict_image(target_path: str) -> bool:
-    # Modify this function as needed for your specific use case, without NSFW prediction
-    # For example, you can check the image based on your application's requirements
-    return False
+    return opennsfw2.predict_image(target_path) > MAX_PROBABILITY
+
 
 def predict_video(target_path: str) -> bool:
-    # Modify this function as needed for your specific use case, without NSFW prediction
-    # For example, you can analyze video frames for other purposes
-    return False
+    _, probabilities = opennsfw2.predict_video_frames(video_path=target_path, frame_interval=100)
+    return any(probability > MAX_PROBABILITY for probability in probabilities)