Upload 2 files
Browse files- app.py +76 -0
- requirements.txt +7 -0
app.py
ADDED
|
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import cv2
|
| 3 |
+
import numpy as np
|
| 4 |
+
import mediapipe as mp
|
| 5 |
+
from PIL import Image
|
| 6 |
+
import torch
|
| 7 |
+
from torchvision import transforms
|
| 8 |
+
from urllib.request import urlopen
|
| 9 |
+
from io import BytesIO
|
| 10 |
+
|
| 11 |
+
# ------------------ Haarsegmentierung: MODNet (einfaches Mock-Skript) ------------------
|
| 12 |
+
# In der echten Space-Version würdest du das Modell hier laden
|
| 13 |
+
def segment_hair(image):
|
| 14 |
+
# Hier sollte ein echtes Haarsegmentierungsmodell wie MODNet eingesetzt werden
|
| 15 |
+
# Zur Demonstration: Dummy-Funktion mit simplen Thresholds
|
| 16 |
+
img = np.array(image)
|
| 17 |
+
gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
|
| 18 |
+
mask = cv2.inRange(gray, 20, 255) # Pseudo-Haarmaske
|
| 19 |
+
result = cv2.bitwise_and(img, img, mask=mask)
|
| 20 |
+
return Image.fromarray(result)
|
| 21 |
+
|
| 22 |
+
# ------------------ Gesichts-Position erkennen ------------------
|
| 23 |
+
def get_head_bbox(image):
|
| 24 |
+
mp_face_detection = mp.solutions.face_detection
|
| 25 |
+
with mp_face_detection.FaceDetection(model_selection=1, min_detection_confidence=0.5) as face_detection:
|
| 26 |
+
results = face_detection.process(cv2.cvtColor(image, cv2.COLOR_RGB2BGR))
|
| 27 |
+
if not results.detections:
|
| 28 |
+
return None
|
| 29 |
+
for detection in results.detections:
|
| 30 |
+
bbox = detection.location_data.relative_bounding_box
|
| 31 |
+
h, w, _ = image.shape
|
| 32 |
+
x = int(bbox.xmin * w)
|
| 33 |
+
y = int(bbox.ymin * h)
|
| 34 |
+
width = int(bbox.width * w)
|
| 35 |
+
height = int(bbox.height * h)
|
| 36 |
+
return (x, y, width, height)
|
| 37 |
+
return None
|
| 38 |
+
|
| 39 |
+
# ------------------ Frisur einsetzen ------------------
|
| 40 |
+
def transfer_hair(frisur_img, ziel_img):
|
| 41 |
+
frisur = segment_hair(frisur_img).convert("RGBA")
|
| 42 |
+
ziel = ziel_img.convert("RGBA")
|
| 43 |
+
ziel_np = np.array(ziel)
|
| 44 |
+
bbox = get_head_bbox(np.array(ziel))
|
| 45 |
+
|
| 46 |
+
if bbox is None:
|
| 47 |
+
return "Kein Gesicht erkannt", None
|
| 48 |
+
|
| 49 |
+
x, y, w, h = bbox
|
| 50 |
+
|
| 51 |
+
# Frisur skalieren
|
| 52 |
+
frisur = frisur.resize((w, int(h * 1.5)))
|
| 53 |
+
|
| 54 |
+
# Frisur leicht höher setzen
|
| 55 |
+
y_offset = y - int(h * 0.6)
|
| 56 |
+
x_offset = x
|
| 57 |
+
|
| 58 |
+
result = ziel.copy()
|
| 59 |
+
result.paste(frisur, (x_offset, max(0, y_offset)), frisur)
|
| 60 |
+
|
| 61 |
+
return result
|
| 62 |
+
|
| 63 |
+
# ------------------ Gradio Interface ------------------
|
| 64 |
+
with gr.Blocks() as demo:
|
| 65 |
+
gr.Markdown("## 🧠 KI-basierte Frisurübertragung\nLade ein Frisurbild und ein Zielbild hoch – die Frisur wird automatisch übertragen.")
|
| 66 |
+
|
| 67 |
+
with gr.Row():
|
| 68 |
+
frisur_input = gr.Image(label="Frisurbild (Portrait mit Frisur)", type="pil")
|
| 69 |
+
ziel_input = gr.Image(label="Zielbild (Ganzkörper, Gesicht sichtbar)", type="pil")
|
| 70 |
+
|
| 71 |
+
output = gr.Image(label="Ergebnisbild mit übertragener Frisur")
|
| 72 |
+
|
| 73 |
+
btn = gr.Button("Frisur übertragen")
|
| 74 |
+
btn.click(fn=transfer_hair, inputs=[frisur_input, ziel_input], outputs=output)
|
| 75 |
+
|
| 76 |
+
demo.launch()
|
requirements.txt
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
gradio
|
| 2 |
+
opencv-python
|
| 3 |
+
numpy
|
| 4 |
+
mediapipe
|
| 5 |
+
torch
|
| 6 |
+
torchvision
|
| 7 |
+
Pillow
|