cut face

app.py

@@ -23,6 +23,10 @@ import skimage
 import mediapipe as mp
 import numpy as np
 
+#import pathlib
+#temp = pathlib.PosixPath
+#pathlib.PosixPath = pathlib.WindowsPath
+
 from huggingface_hub import hf_hub_download
 ckpt = hf_hub_download(repo_id="lambdalabs/image-mixer", filename="image-mixer-pruned.ckpt", cache_dir="/data/.cache")
 config = hf_hub_download(repo_id="lambdalabs/image-mixer", filename="image-mixer-config.yaml", cache_dir="/data/.cache")
@@ -141,13 +145,51 @@ def ethnicity(img):
     pred,pred_idx,probs = ethnic_learn.predict(img)
     return pred
 
-
-import gradio
+def find_face(img):
+    mp_face_detection = mp.solutions.face_detection
+    img_array = np.array(img)
+    print(img_array.shape)
+    
+    with mp_face_detection.FaceDetection(
+            model_selection=0,
+            min_detection_confidence=0.75
+    ) as face_detection:
+        results = face_detection.process(img_array)
+        
+    if results.detections is not None:
+        return results.detections[0]
+    
+    return None
 
 def boutsify(person):
+    detected_face = find_face(person)
+    
+    if detected_face is None:
+        print("Couldn't detect a face")        
+        return
+    
+    person_image = Image.fromarray(person)
+    bounding_box = detected_face.location_data.relative_bounding_box
+    width, height = person_image.size
+    
+    width_margin = bounding_box.width * 0.5
+    height_margin = bounding_box.height * 0.8
+    # Setting the points for cropped image
+    left = max(0, (bounding_box.xmin - width_margin / 2.0) * width)
+    top = max(0, (bounding_box.ymin - height_margin / 1.2) * height)
+    
+    right = min(width, left + (bounding_box.width + width_margin) * width)
+    bottom = min(height, top + (bounding_box.height + height_margin) * height)
+     
+    # Cropped image of above dimension
+    # (It will not change original image)
+    face = person_image.crop((left, top, right, bottom))
+    face_array = np.array(face)
+    
+
     portrait_path = "bouts_m1.jpg"
-    female_detected = is_female(person)
-    ethnicity_prediction = ethnicity(person)
+    female_detected = is_female(face_array)
+    ethnicity_prediction = ethnicity(face_array)
     
     if ethnicity_prediction == "Black":
         print("Colored person")
@@ -162,19 +204,17 @@ def boutsify(person):
             portrait_path = "bouts_f1.jpg"
         else:
             print("This is a male portrait, checking facial hair")
-            if has_beard(person):
+            if has_beard(face_array):
                 print("The person has a beard")
                 portrait_path = "bouts_mb1.jpg"
     
-    person_image = Image.fromarray(person)
-    
     inputs = [
         "Image", "Image", "Image", "Image", "Nothing",
         "","","","","",
         Image.open(portrait_path).convert("RGB"),
         Image.open("boutsback_o1.png").convert("RGB"),
         Image.open("boutsback_o2.png").convert("RGB"),
-        person_image,
+        face,
         "",
         1.1,1,1,1.4,1,
         3.0, 1, random.randrange(0, 10000), 50,
@@ -182,6 +222,9 @@ def boutsify(person):
     
     #return person
     return run_image_mixer(inputs)
+    
+    
+import gradio
 
 gradio_interface = gradio.Interface(
     fn=boutsify,