Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,49 +1,92 @@
|
|
1 |
import os
|
2 |
-
import
|
3 |
-
from PIL import Image
|
4 |
import io
|
5 |
-
|
6 |
-
from
|
|
|
7 |
from google.genai import types
|
8 |
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
-
|
11 |
-
|
12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
|
|
14 |
st.title("Bildanalyse mit Gemini")
|
15 |
col1, col2 = st.columns(2)
|
16 |
|
17 |
with col1:
|
18 |
-
|
19 |
uploaded_file = st.file_uploader("Bild hochladen", type=["jpg", "png", "jpeg"])
|
20 |
-
|
21 |
-
|
|
|
22 |
image = Image.open(uploaded_file)
|
23 |
st.image(image, caption="Hochgeladenes Bild", use_container_width=True)
|
24 |
-
|
25 |
if st.button("Analysieren"):
|
26 |
with st.spinner("Analysiere Bild..."):
|
27 |
try:
|
28 |
-
#
|
29 |
image_bytes = io.BytesIO()
|
30 |
image.save(image_bytes, format=image.format)
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
# Anfrage an Gemini senden
|
35 |
-
client = genai.Client(api_key=os.getenv("KEY")) # Client innerhalb der Funktion erstellen
|
36 |
-
response = client.models.generate_content(
|
37 |
-
model="gemini-2.0-flash-exp", # Oder "gemini-2.0-flash-exp", je nach Verfügbarkeit
|
38 |
-
contents=["Beschreibe dieses Bild und identifiziere das Hauptobjekt.", types.Part.from_bytes(data=image_bytes, mime_type=f"image/{image.format.lower()}")
|
39 |
-
]
|
40 |
)
|
41 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
42 |
with col2:
|
43 |
-
|
44 |
-
st.write(
|
45 |
-
|
|
|
|
|
|
|
46 |
|
47 |
-
|
48 |
except Exception as e:
|
49 |
-
st.error(f"
|
|
|
1 |
import os
|
2 |
+
import re
|
|
|
3 |
import io
|
4 |
+
import streamlit as st
|
5 |
+
from PIL import Image, ImageDraw
|
6 |
+
from google import genai
|
7 |
from google.genai import types
|
8 |
|
9 |
+
# Hilfsfunktionen
|
10 |
+
def parse_list_boxes(text):
|
11 |
+
"""Extrahiert Bounding Boxes aus dem Antworttext"""
|
12 |
+
pattern = r'\[([\d\.]+),\s*([\d\.]+),\s*([\d\.]+),\s*([\d\.]+)\]'
|
13 |
+
matches = re.findall(pattern, text)
|
14 |
+
return [[float(m) for m in match] for match in matches]
|
15 |
|
16 |
+
def draw_bounding_boxes(image, boxes):
|
17 |
+
"""Zeichnet Bounding Boxes auf das Bild"""
|
18 |
+
draw = ImageDraw.Draw(image)
|
19 |
+
width, height = image.size
|
20 |
+
for box in boxes:
|
21 |
+
ymin, xmin, ymax, xmax = box
|
22 |
+
draw.rectangle([
|
23 |
+
xmin * width,
|
24 |
+
ymin * height,
|
25 |
+
xmax * width,
|
26 |
+
ymax * height
|
27 |
+
], outline="red", width=3)
|
28 |
+
return image
|
29 |
|
30 |
+
# Streamlit UI
|
31 |
st.title("Bildanalyse mit Gemini")
|
32 |
col1, col2 = st.columns(2)
|
33 |
|
34 |
with col1:
|
|
|
35 |
uploaded_file = st.file_uploader("Bild hochladen", type=["jpg", "png", "jpeg"])
|
36 |
+
object_name = st.text_input("Objekt zur Erkennung", placeholder="z.B. 'Auto', 'Person'")
|
37 |
+
|
38 |
+
if uploaded_file and object_name:
|
39 |
image = Image.open(uploaded_file)
|
40 |
st.image(image, caption="Hochgeladenes Bild", use_container_width=True)
|
41 |
+
|
42 |
if st.button("Analysieren"):
|
43 |
with st.spinner("Analysiere Bild..."):
|
44 |
try:
|
45 |
+
# Bildvorbereitung
|
46 |
image_bytes = io.BytesIO()
|
47 |
image.save(image_bytes, format=image.format)
|
48 |
+
image_part = types.Part.from_bytes(
|
49 |
+
data=image_bytes.getvalue(),
|
50 |
+
mime_type=f"image/{image.format.lower()}"
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
)
|
52 |
+
|
53 |
+
# API-Client
|
54 |
+
client = genai.Client(api_key=os.getenv("KEY"))
|
55 |
+
|
56 |
+
# Bildbeschreibung
|
57 |
+
desc_response = client.models.generate_content(
|
58 |
+
model="gemini-1.0-pro-vision",
|
59 |
+
contents=["Beschreibe dieses Bild detailliert.", image_part]
|
60 |
+
)
|
61 |
+
|
62 |
+
# Objekterkennung
|
63 |
+
detection_prompt = (
|
64 |
+
f"Gib alle Bounding Boxes für {object_name} im Format "
|
65 |
+
"[ymin, xmin, ymax, xmax] als Liste. Nur die Liste zurückgeben!"
|
66 |
+
)
|
67 |
+
box_response = client.models.generate_content(
|
68 |
+
model="gemini-1.0-pro-vision",
|
69 |
+
contents=[detection_prompt, image_part]
|
70 |
+
)
|
71 |
+
|
72 |
+
# Verarbeitung
|
73 |
+
boxes = parse_list_boxes(box_response.text)
|
74 |
+
annotated_image = image.copy()
|
75 |
+
|
76 |
+
if boxes:
|
77 |
+
annotated_image = draw_bounding_boxes(annotated_image, boxes)
|
78 |
+
result_text = f"{len(boxes)} {object_name} erkannt"
|
79 |
+
else:
|
80 |
+
result_text = "Keine Objekte gefunden"
|
81 |
+
|
82 |
+
# Ergebnisse anzeigen
|
83 |
with col2:
|
84 |
+
st.write("## Beschreibung:")
|
85 |
+
st.write(desc_response.text)
|
86 |
+
|
87 |
+
st.write("## Objekterkennung:")
|
88 |
+
st.write(result_text)
|
89 |
+
st.image(annotated_image, caption="Erkannte Objekte", use_column_width=True)
|
90 |
|
|
|
91 |
except Exception as e:
|
92 |
+
st.error(f"Fehler: {str(e)}")
|