init

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.task filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1 @@
+__pycache__

README.md

@@ -1,5 +1,5 @@
 ---
-title: Face Crop And Replace CPU
+title: mediapipe-face-crop-and-replace
 emoji: 📈
 colorFrom: indigo
 colorTo: indigo

app.py

@@ -0,0 +1,231 @@
+import spaces
+import gradio as gr
+import subprocess
+from PIL import Image,ImageEnhance,ImageFilter
+import json
+
+import mp_box
+'''
+Face landmark detection based Face Detection.
+https://ai.google.dev/edge/mediapipe/solutions/vision/face_landmarker
+from model card
+https://storage.googleapis.com/mediapipe-assets/MediaPipe%20BlazeFace%20Model%20Card%20(Short%20Range).pdf
+Licensed Apache License, Version 2.0
+Train with google's dataset(more detail see model card)
+
+Not Face Detector based 
+https://ai.google.dev/edge/mediapipe/solutions/vision/face_detector
+
+Bacause this is part of getting-landmark program and need control face edge.
+So I don't know which one is better.never compare these.
+'''
+
+def select_box(boxes,box_type):
+    if box_type == "type-3":
+         box = boxes[2]
+    elif box_type =="type-2":
+         box = boxes[1]
+    elif box_type =="type-1":
+         box = boxes[0]
+    else:#never happen
+         box=[0,0,image.size[0],image.size[1]]
+    box_width = box[2]
+    box_height = box[3]
+    box = mp_box.xywh_to_xyxy(box)
+    return box,box_width,box_height
+
+def process_images(image,replace_image=None,replace_image_need_crop=False,box_type="type-3",fill_color_mode=False,fill_color="black",custom_color="rgba(255,255,255,1)",image_size=1024,filter_image=False,filter_value="Sharpen",progress=gr.Progress(track_tqdm=True)):
+    if image == None:
+        raise gr.Error("Need Image")
+
+    # choose box
+    boxes,mp_image,face_landmarker_result = mp_box.mediapipe_to_box(image)
+    box,box_width,box_height = select_box(boxes,box_type)
+
+    # replace-mode
+    if replace_image!=None:
+        print("replace mode")
+
+        if replace_image_need_crop:
+            replace_boxes,mp_image,face_landmarker_result = mp_box.mediapipe_to_box(replace_image)
+            replace_box,replace_box_width,replace_box_height = select_box(replace_boxes,box_type)
+
+        if fill_color_mode:
+            if replace_image_need_crop:
+                cropped = replace_image.crop(replace_box)
+                cropped.resize(box_width,box_height)
+            else:
+                cropped = replace_image.crop(box)
+            image.paste(cropped,[box[0],box[1]])
+            return image
+        else:#scale mode
+            if replace_image_need_crop:
+                replace_image = replace_image.crop(replace_box)
+            replace_resized = replace_image.resize((box_width,box_height),Image.Resampling.LANCZOS)
+            image.paste(replace_resized,[box[0],box[1]])
+            return image
+
+
+    # crop-mode
+    if fill_color_mode:
+        # choose color
+        color_map={
+            "black":[0,0,0,1],
+            "white":[255,255,255,1],
+            "red":[255,0,0,1],
+            "brown":[92,33,31,1],
+            "pink":[255,192,203,1],
+        }
+        if fill_color == "custom":
+            color_value = custom_color.strip("rgba()").split(",")
+            color_value[0] = int(float(color_value[0]))
+            color_value[1] = int(float(color_value[1]))
+            color_value[2] = int(float(color_value[2]))
+        else:
+            color_value = color_map[fill_color]
+
+        cropped = image.crop(box)
+
+        img = Image.new('RGBA', image.size, (color_value[0], color_value[1], color_value[2]))
+        img.paste(cropped,[box[0],box[1]])
+        return img
+    else:
+        #scale up mode
+        cropped = image.crop(box)
+        resized = resize_image_by_max_dimension(cropped,image_size)
+        filter_map={
+           "Blur":ImageFilter.BLUR,"Smooth More":ImageFilter.SMOOTH_MORE,"Smooth":ImageFilter.SMOOTH,"Sharpen":ImageFilter.SHARPEN,"Edge Enhance":ImageFilter.EDGE_ENHANCE,"Edge Enhance More":ImageFilter.EDGE_ENHANCE_MORE
+        }
+        if filter_value not in filter_map:
+            raise gr.Error(f"filter {filter_value} not found")
+        if filter_image:
+           
+            #resized = resized.filter(ImageFilter.SHARPEN)
+            #Gimp's weak 0.1-0.2?
+            enhancer = ImageEnhance.Sharpness(resized)
+            resized = resized.filter(filter_map[filter_value])
+            #resized = enhancer.enhance(sharpen_value)
+
+
+        return resized
+
+
+def resize_image_by_max_dimension(image, max_size, resampling=Image.Resampling.BICUBIC):
+    image_width, image_height = image.size
+    
+    max_dimension = max(image_width, image_height)
+    
+    ratio = max_size / max_dimension
+    
+    new_width = int(image_width * ratio)
+    new_height = int(image_height * ratio)
+ 
+    return image.resize((new_width, new_height), resampling)
+    
+
+def read_file(file_path: str) -> str:
+    """read the text of target file
+    """
+    with open(file_path, 'r', encoding='utf-8') as f:
+        content = f.read()
+
+    return content
+
+css="""
+#col-left {
+    margin: 0 auto;
+    max-width: 640px;
+}
+#col-right {
+    margin: 0 auto;
+    max-width: 640px;
+}
+.grid-container {
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  gap:10px
+}
+
+.image {
+  width: 128px; 
+  height: 128px; 
+  object-fit: cover; 
+}
+
+.text {
+  font-size: 16px;
+}
+"""
+
+#css=css,
+def update_button_label(image):
+    if image == None:
+        return  gr.Button(visible=bool(0)),gr.Button(visible=bool(1))
+    else:
+        return  gr.Button(visible=bool(1)),gr.Row(visible=bool(0))   
+    
+def update_visible(fill_color_mode):
+     if fill_color_mode:
+        return  gr.Row(visible=bool(0)),gr.Row(visible=bool(1))
+     else:
+        return  gr.Row(visible=bool(1)),gr.Row(visible=bool(0))    
+with gr.Blocks(css=css, elem_id="demo-container") as demo:
+    with gr.Column():
+        gr.HTML(read_file("demo_header.html"))
+        gr.HTML(read_file("demo_tools.html"))
+    with gr.Row():
+                with gr.Column():
+                    image = gr.Image(sources=['upload','clipboard'],image_mode='RGB',elem_id="image_upload", type="pil", label="Upload")
+                    box_type = gr.Dropdown(label="box-type",value="type-3",choices=["type-1","type-2","type-3"])
+                    with gr.Row(elem_id="prompt-container",  equal_height=False):
+                        with gr.Row():
+                            btn1 = gr.Button("Face Crop", elem_id="run_button",variant="primary")
+                            btn2 = gr.Button("Face Replace", elem_id="run_button2",variant="primary",visible=False)
+                            
+                    replace_image = gr.Image(sources=['upload','clipboard'],image_mode='RGB',elem_id="replace_upload", type="pil", label="replace image")
+                    replace_image_need_crop = gr.Checkbox(label="Replace image need crop",value=False)
+                    replace_image.change(update_visible,replace_image,[btn1,btn2])
+                    with gr.Accordion(label="Advanced Settings", open=False):
+                        fill_color_mode = gr.Checkbox(label="Fill Color Mode",value=False)
+                        row1 = gr.Row(equal_height=True)
+                        row2 = gr.Row(equal_height=True,visible=False)
+                        fill_color_mode.change(update_visible,fill_color_mode,[row1,row2])
+                        
+                        with row1:
+                            image_size = gr.Slider(
+                            label="Image Size",info = "cropped face size",
+                            minimum=8,
+                            maximum=2048,
+                            step=1,
+                            value=1024,
+                            interactive=True)
+
+                            filter_image = gr.Checkbox(label="Filter image")
+                            filter_value = gr.Dropdown(label="Filter",value="Sharpen",choices=["Blur","Smooth More","Smooth","Sharpen","Edge Enhance","Edge Enhance More"])
+                        with row2:
+                            
+                            fill_color = gr.Dropdown(label="fill color",choices=["black","white","red","brown","pink","custom"])
+                            custom_color = gr.ColorPicker(label="custom color",value="rgba(250, 218, 205, 1)")
+                            
+                    
+                with gr.Column():
+                    image_out = gr.Image(label="Output", elem_id="output-img")
+                    
+                    
+                    
+            
+
+    
+    gr.on(
+        [btn1.click,btn2.click],
+        fn=process_images, inputs=[image,replace_image,replace_image_need_crop,box_type,fill_color_mode,fill_color,custom_color,image_size,filter_image,filter_value], outputs =[image_out], api_name='infer'
+    )
+    gr.Examples(
+                examples =["examples/00004200.jpg"],
+                inputs=[image]
+    )
+    gr.HTML(read_file("demo_footer.html"))
+
+    if __name__ == "__main__":
+        demo.launch()

demo_footer.html

@@ -0,0 +1,3 @@
+<div>
+    <P> Images are generated with <a href="https://huggingface.co/black-forest-labs/FLUX.1-schnell">FLUX.1-schnell</a> and licensed under <a href="http://www.apache.org/licenses/LICENSE-2.0">the Apache 2.0 License</a>
+</div>

demo_header.html

@@ -0,0 +1,15 @@
+<div style="text-align: center;">
+    <h1>
+        Mediapipe Face Crop and Replace
+    </h1>
+    <div  class="grid-container">
+        <img src="https://akjava.github.io/AIDiagramChatWithVoice-FaceCharacter/webp/128/00191245_09_00002200.webp" alt="Mediapipe Face Detection" class="image">
+       
+        <p class="text">
+            This Space use <a href="http://www.apache.org/licenses/LICENSE-2.0">the Apache 2.0</a> Licensed <a href="https://ai.google.dev/edge/mediapipe/solutions/vision/face_landmarker">Mediapipe FaceLandmarker</a> <br>
+
+            
+        </p>
+    </div>
+    
+</div>

demo_tools.html

@@ -0,0 +1,4 @@
+<div style="text-align: center;">
+    <p><a href="https://huggingface.co/spaces/Akjava/mediapipe-face-detect">Mediapipe Face detector</a></p>
+    <p></p> 
+</div>

face_landmarker.task

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64184e229b263107bc2b804c6625db1341ff2bb731874b0bcc2fe6544e0bc9ff
+size 3758596

face_landmarker.task.txt

@@ -0,0 +1,8 @@
+Face landmark detection
+https://ai.google.dev/edge/mediapipe/solutions/vision/face_landmarker
+
+model card page is 
+https://storage.googleapis.com/mediapipe-assets/MediaPipe%20BlazeFace%20Model%20Card%20(Short%20Range).pdf
+
+license is Apache2.0
+https://www.apache.org/licenses/LICENSE-2.0.html

mp_box.py

@@ -0,0 +1,133 @@
+import mediapipe as mp
+from mediapipe.tasks import python
+from mediapipe.tasks.python import vision
+from mediapipe.framework.formats import landmark_pb2
+from mediapipe import solutions
+import numpy as np
+
+# for X,Y,W,H to x1,y1,x2,y2(Left-top,right-bottom style)
+def xywh_to_xyxy(box):
+  return [box[0],box[1],box[0]+box[2],box[1]+box[3]]
+
+def convert_to_box(face_landmarks_list,indices,w=1024,h=1024):
+  x1=w
+  y1=h
+  x2=0
+  y2=0
+  for index in indices:
+    x=min(w,max(0,(face_landmarks_list[0][index].x*w)))
+    y=min(h,max(0,(face_landmarks_list[0][index].y*h)))
+    if x<x1:
+      x1=x
+
+    if y<y1:
+      y1=y
+    
+    if x>x2:
+      x2=x
+    if y>y2:
+      y2=y
+   
+        
+  return [int(x1),int(y1),int(x2-x1),int(y2-y1)]
+       
+  
+def box_to_square(bbox):
+  box=list(bbox)
+  if box[2]>box[3]:
+    diff = box[2]-box[3]
+    box[3]+=diff
+    box[1]-=diff/2
+  elif box[3]>box[2]:
+    diff = box[3]-box[2]
+    box[2]+=diff
+    box[0]-=diff/2
+  return box
+
+
+def face_landmark_result_to_box(face_landmarker_result,width=1024,height=1024):
+  face_landmarks_list = face_landmarker_result.face_landmarks
+
+
+  full_indices  = list(range(456))
+
+  MIDDLE_FOREHEAD = 151
+  BOTTOM_CHIN_EX = 152
+  BOTTOM_CHIN = 175
+  CHIN_TO_MIDDLE_FOREHEAD = [200,14,1,6,18,9]
+  MOUTH_BOTTOM = [202,200,422]
+  EYEBROW_CHEEK_LEFT_RIGHT = [46,226,50,1,280,446,276]
+
+  LEFT_HEAD_OUTER_EX = 251  #on side face almost same as full
+  LEFT_HEAD_OUTER = 301
+  LEFT_EYE_OUTER_EX = 356
+  LEFT_EYE_OUTER = 264
+  LEFT_MOUTH_OUTER_EX = 288
+  LEFT_MOUTH_OUTER = 288
+  LEFT_CHIN_OUTER = 435
+  RIGHT_HEAD_OUTER_EX = 21
+  RIGHT_HEAD_OUTER = 71
+  RIGHT_EYE_OUTER_EX = 127
+  RIGHT_EYE_OUTER = 34
+  RIGHT_MOUTH_OUTER_EX = 58
+  RIGHT_MOUTH_OUTER = 215
+  RIGHT_CHIN_OUTER = 150
+
+  # TODO naming line
+  min_indices=CHIN_TO_MIDDLE_FOREHEAD+EYEBROW_CHEEK_LEFT_RIGHT+MOUTH_BOTTOM
+
+  chin_to_brow_indices = [LEFT_CHIN_OUTER,LEFT_MOUTH_OUTER,LEFT_EYE_OUTER,LEFT_HEAD_OUTER,MIDDLE_FOREHEAD,RIGHT_HEAD_OUTER,RIGHT_EYE_OUTER,RIGHT_MOUTH_OUTER,RIGHT_CHIN_OUTER,BOTTOM_CHIN]+min_indices
+  
+  box1 = convert_to_box(face_landmarks_list,min_indices,width,height)
+  box2 = convert_to_box(face_landmarks_list,chin_to_brow_indices,width,height)
+  box3 = convert_to_box(face_landmarks_list,full_indices,width,height)
+  #print(box)
+
+  return [box1,box2,box3,box_to_square(box1),box_to_square(box2),box_to_square(box3)]
+
+
+def draw_landmarks_on_image(detection_result,rgb_image):
+  face_landmarks_list = detection_result.face_landmarks
+  annotated_image = np.copy(rgb_image)
+
+  # Loop through the detected faces to visualize.
+  for idx in range(len(face_landmarks_list)):
+    face_landmarks = face_landmarks_list[idx]
+
+    # Draw the face landmarks.
+    face_landmarks_proto = landmark_pb2.NormalizedLandmarkList()
+    face_landmarks_proto.landmark.extend([
+      landmark_pb2.NormalizedLandmark(x=landmark.x, y=landmark.y, z=landmark.z) for landmark in face_landmarks
+    ])
+
+    solutions.drawing_utils.draw_landmarks(
+        image=annotated_image,
+        landmark_list=face_landmarks_proto,
+        connections=mp.solutions.face_mesh.FACEMESH_TESSELATION,
+        landmark_drawing_spec=None,
+        connection_drawing_spec=mp.solutions.drawing_styles
+        .get_default_face_mesh_tesselation_style())
+    
+  return annotated_image
+
+def mediapipe_to_box(image_data,model_path="face_landmarker.task"):
+  BaseOptions = mp.tasks.BaseOptions
+  FaceLandmarker = mp.tasks.vision.FaceLandmarker
+  FaceLandmarkerOptions = mp.tasks.vision.FaceLandmarkerOptions
+  VisionRunningMode = mp.tasks.vision.RunningMode
+
+  options = FaceLandmarkerOptions(
+      base_options=BaseOptions(model_asset_path=model_path),
+      running_mode=VisionRunningMode.IMAGE
+      ,min_face_detection_confidence=0, min_face_presence_confidence=0
+      )
+
+
+  with FaceLandmarker.create_from_options(options) as landmarker:
+    if isinstance(image_data,str):
+        mp_image = mp.Image.create_from_file(image_data)
+    else:
+        mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=np.asarray(image_data))
+    face_landmarker_result = landmarker.detect(mp_image)
+    boxes = face_landmark_result_to_box(face_landmarker_result,mp_image.width,mp_image.height)
+    return boxes,mp_image,face_landmarker_result

requirements.txt

@@ -0,0 +1,4 @@
+numpy
+torch
+spaces
+mediapipe