Spaces:
Runtime error
Runtime error
syurein
commited on
Commit
·
2443b6b
1
Parent(s):
fd6583c
jinja修正
Browse files- LLM_package.py +103 -0
- app.py +14 -4
LLM_package.py
ADDED
|
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from google import genai
|
| 2 |
+
import moondream as md
|
| 3 |
+
import json
|
| 4 |
+
import os
|
| 5 |
+
from PIL import Image
|
| 6 |
+
from dotenv import load_dotenv
|
| 7 |
+
load_dotenv()
|
| 8 |
+
class MoondreamInference:
|
| 9 |
+
def __init__(self, api_key=None):
|
| 10 |
+
if api_key is None:
|
| 11 |
+
api_key = os.getenv('MOONDREAM_API_KEY')
|
| 12 |
+
self.model = md.vl(api_key=api_key)
|
| 13 |
+
|
| 14 |
+
def get_response(self, image_path, prompt):
|
| 15 |
+
"""
|
| 16 |
+
COCOEvaluator は get_response を呼ぶので、
|
| 17 |
+
ここで Moondream の detect を内部で呼び、結果を JSON文字列で返す
|
| 18 |
+
"""
|
| 19 |
+
image = Image.open(image_path)
|
| 20 |
+
cat = list(prompt) # prompt を直接カテゴリ名に使う
|
| 21 |
+
result = self.model.detect(image, list(cat)[0])
|
| 22 |
+
# Moondream はすでに dict なので JSON にして返す
|
| 23 |
+
return json.dumps(result["objects"])
|
| 24 |
+
|
| 25 |
+
def parse_response(self, resp_text):
|
| 26 |
+
"""
|
| 27 |
+
get_response で返した JSON文字列をパースし、
|
| 28 |
+
Gemini と同じ形式の list[dict] に揃える
|
| 29 |
+
"""
|
| 30 |
+
|
| 31 |
+
detections = json.loads(resp_text)
|
| 32 |
+
parsed = []
|
| 33 |
+
for obj in detections:
|
| 34 |
+
parsed.append({
|
| 35 |
+
"label": obj.get("label", "object"), # ない場合もあるかも
|
| 36 |
+
"box_2d": [
|
| 37 |
+
obj["y_min"], obj["x_min"],
|
| 38 |
+
obj["y_max"], obj["x_max"]
|
| 39 |
+
]
|
| 40 |
+
})
|
| 41 |
+
print(parsed)
|
| 42 |
+
return parsed
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
class GeminiInference:
|
| 46 |
+
"""
|
| 47 |
+
Gemini API 呼び出しを扱うクラス。
|
| 48 |
+
"""
|
| 49 |
+
def __init__(self, api_key_source=os.getenv('GEMINI_API_KEY')):
|
| 50 |
+
self.api_key_source = api_key_source
|
| 51 |
+
|
| 52 |
+
def get_response(self, file_path, prompt):
|
| 53 |
+
"""
|
| 54 |
+
画像ファイルに対して Geminin API 呼び出しを行い、レスポンステキストを返す。
|
| 55 |
+
"""
|
| 56 |
+
client = genai.Client(api_key=self.api_key_source)
|
| 57 |
+
my_file = client.files.upload(file=file_path)
|
| 58 |
+
response = client.models.generate_content(
|
| 59 |
+
model="gemini-2.0-flash",
|
| 60 |
+
contents=[my_file, prompt],
|
| 61 |
+
)
|
| 62 |
+
return response.text
|
| 63 |
+
def get_response_text(self,prompt):
|
| 64 |
+
client = genai.Client(api_key=self.api_key_source)
|
| 65 |
+
response = client.models.generate_content(
|
| 66 |
+
model="gemini-2.0-flash",
|
| 67 |
+
contents=[prompt],
|
| 68 |
+
)
|
| 69 |
+
text = response.text
|
| 70 |
+
return text
|
| 71 |
+
def parse(self, text):
|
| 72 |
+
"""
|
| 73 |
+
レスポンス JSON をパース。'label' と 'box_2d'([0-1000]正規化) を取り出し、[0,1]正規化に変換して返すリスト。
|
| 74 |
+
"""
|
| 75 |
+
json_str = text
|
| 76 |
+
if '```json' in text:
|
| 77 |
+
json_str = text[text.find('```json') + len('```json'):]
|
| 78 |
+
|
| 79 |
+
json_str = json_str.strip('` \n')
|
| 80 |
+
return json_str
|
| 81 |
+
def parse_response(self, text):
|
| 82 |
+
"""
|
| 83 |
+
レスポンス JSON をパース。'label' と 'box_2d'([0-1000]正規化) を取り出し、[0,1]正規化に変換して返すリスト。
|
| 84 |
+
"""
|
| 85 |
+
print(text)
|
| 86 |
+
json_str = text
|
| 87 |
+
if '```json' in text:
|
| 88 |
+
json_str = text[text.find('```json') + len('```json'):]
|
| 89 |
+
json_str = json_str.strip('` \n')
|
| 90 |
+
try:
|
| 91 |
+
data = json.loads(json_str)
|
| 92 |
+
except Exception as e:
|
| 93 |
+
print("JSON パースエラー:", e)
|
| 94 |
+
return []
|
| 95 |
+
if isinstance(data, dict):
|
| 96 |
+
data = [data]
|
| 97 |
+
parsed = []
|
| 98 |
+
for obj in data:
|
| 99 |
+
if 'box_2d' in obj and 'label' in obj:
|
| 100 |
+
coords = obj['box_2d']
|
| 101 |
+
norm = [c / 1000.0 for c in coords]
|
| 102 |
+
parsed.append({'label': obj['label'], 'box_2d': norm})
|
| 103 |
+
return parsed
|
app.py
CHANGED
|
@@ -50,7 +50,8 @@ from ultralytics import YOLO
|
|
| 50 |
import math
|
| 51 |
import numpy as np
|
| 52 |
import matplotlib.pyplot as plt
|
| 53 |
-
|
|
|
|
| 54 |
#この下のコードは特定の領域をマスクしないタイプのコード
|
| 55 |
import uuid
|
| 56 |
from datetime import datetime
|
|
@@ -58,7 +59,7 @@ import torch
|
|
| 58 |
import cv2
|
| 59 |
import numpy as np
|
| 60 |
from ultralytics import YOLO # YOLOv8ライブラリ
|
| 61 |
-
|
| 62 |
import random
|
| 63 |
import cv2
|
| 64 |
import numpy as np
|
|
@@ -75,7 +76,7 @@ app.add_middleware(
|
|
| 75 |
allow_headers=["*"],
|
| 76 |
)
|
| 77 |
|
| 78 |
-
|
| 79 |
HOME = "./"
|
| 80 |
templates = Jinja2Templates(directory="templates")
|
| 81 |
dangerarray=[10,30,90,50,80,20,40,70,100,60]#ここに各クラスターの危険度を設定しておく
|
|
@@ -224,6 +225,13 @@ def create_mask(image, x1, y1, x2, y2):
|
|
| 224 |
|
| 225 |
import easyocr
|
| 226 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 227 |
# 特殊な処理を行う関数
|
| 228 |
def special_process_image_yolo(risk_level, image_path, point1, point2, thresholds=None):
|
| 229 |
# デバイスの確認
|
|
@@ -902,5 +910,7 @@ async def mosaic_faces(reference_image: UploadFile = File(...), test_image: Uplo
|
|
| 902 |
|
| 903 |
@app.get("/", response_class=HTMLResponse)
|
| 904 |
async def read_root():
|
| 905 |
-
return templates.TemplateResponse("index.html")
|
|
|
|
|
|
|
| 906 |
|
|
|
|
| 50 |
import math
|
| 51 |
import numpy as np
|
| 52 |
import matplotlib.pyplot as plt
|
| 53 |
+
from dotenv import load_dotenv
|
| 54 |
+
from pathlib import Path
|
| 55 |
#この下のコードは特定の領域をマスクしないタイプのコード
|
| 56 |
import uuid
|
| 57 |
from datetime import datetime
|
|
|
|
| 59 |
import cv2
|
| 60 |
import numpy as np
|
| 61 |
from ultralytics import YOLO # YOLOv8ライブラリ
|
| 62 |
+
from fastapi.middleware.cors import CORSMiddleware, Request
|
| 63 |
import random
|
| 64 |
import cv2
|
| 65 |
import numpy as np
|
|
|
|
| 76 |
allow_headers=["*"],
|
| 77 |
)
|
| 78 |
|
| 79 |
+
load_dotenv(dotenv_path='../.env')
|
| 80 |
HOME = "./"
|
| 81 |
templates = Jinja2Templates(directory="templates")
|
| 82 |
dangerarray=[10,30,90,50,80,20,40,70,100,60]#ここに各クラスターの危険度を設定しておく
|
|
|
|
| 225 |
|
| 226 |
import easyocr
|
| 227 |
|
| 228 |
+
|
| 229 |
+
|
| 230 |
+
def llm_to_process_image(risk_level, image_path, point1, point2, thresholds=None):
|
| 231 |
+
print('point1,point2', point1, point2)
|
| 232 |
+
# 画像処理のロジックをここに追加
|
| 233 |
+
pass
|
| 234 |
+
|
| 235 |
# 特殊な処理を行う関数
|
| 236 |
def special_process_image_yolo(risk_level, image_path, point1, point2, thresholds=None):
|
| 237 |
# デバイスの確認
|
|
|
|
| 910 |
|
| 911 |
@app.get("/", response_class=HTMLResponse)
|
| 912 |
async def read_root():
|
| 913 |
+
return templates.TemplateResponse("index.html", {"request": request})
|
| 914 |
+
|
| 915 |
+
|
| 916 |
|