TIP_CALCULATOR / app.py
kfkas's picture
git push
f6f4388
import json
import os
import shutil
import time
from typing import Tuple
import cv2
import base64
import uuid
import re
from flask import Flask
import gradio as gr
from google import genai
# --- Config 클래슀 (Gemma, GPT4o 제거, Qwen만 μ‚¬μš©) ---
class Config:
"""μ• ν”Œλ¦¬μΌ€μ΄μ…˜ μ„€μ • 및 μƒμˆ˜"""
FOOD_ITEMS = [
{"name": "짜μž₯λ©΄", "image": "images/food1.jpg", "price": 7.00},
{"name": "짬뽕", "image": "images/food2.jpg", "price": 8.50},
{"name": "νƒ•μˆ˜μœ‘", "image": "images/food3.jpg", "price": 15.00},
{"name": "볢음λ°₯", "image": "images/food4.jpg", "price": 7.50},
{"name": "깐풍기", "image": "images/food5.jpg", "price": 18.00},
{"name": "λ§ˆνŒŒλ‘λΆ€", "image": "images/food6.jpg", "price": 12.00},
{"name": "콜라", "image": "images/food6.jpg", "price": 12.00},
{"name": "사이닀", "image": "images/food6.jpg", "price": 12.00},
]
# μ•Œλ¦¬λ°”λ°” Qwen API ν‚€ (기본값은 빈 λ¬Έμžμ—΄)
QWEN_API_KEY = "sk-2424f0bd26d64fe5a7f3a2bd407adc76"
GEMINI_API_key = "AIzaSyCc8lcm2cZo3ZeMgi4QN1IHSvn9BBpLnz4"
DEFAULT_PROMPT_TEMPLATE = (
"""
### Persona ###
You are an expert tip calculation assistant focusing on service quality observed in video captions, and you also consider the user's review, star rating, and recent Google reviews.
Your role is to evaluate all these aspects, assign scores, and calculate the overall average score to determine the appropriate tip percentage, with special handling for ethical violations.
### Task ###
1. **Video Analysis**: Analyze the service quality depicted in the provided **Video Caption(s)**. If multiple captions are given, perform a **holistic evaluation** based on the overall impression conveyed by all captions combined. **Focus on significant staff actions, expressions, and interactions described across the captions.** Assign a single, overall **Video Service Score** out of 100 based on this comprehensive analysis, rather than averaging scores from individual captions.
2. **Bill Amount Determination**: Determine the bill amount by following these steps:
* Use the 'Calculated Subtotal' provided.
3. **Overall Service Quality Evaluation**:
Evaluate the service quality by evenly scoring the following four components, each out of 100:
a) **Video Service Score**: Service quality derived from the holistic analysis of Video Caption(s).
b) **Google Review Score**: Provide an overall analysis of the recent Google reviews and give the general rating score. Highlight any significant social issues mentioned, such as racist comments or discriminatory behavior. The analysis should consider all reviews collectively.
* **Note**: If any review mentions racist, sexist, or any other ethical violations, the Google Review score must automatically be set to **0**.
c) **User Review Score**: The user's review.
d) **Star Rating Score**: The user's star rating, interpreted on a scale where 5/5 corresponds to 100 points.
* Calculate the overall average score by taking the mean of these four scores.
4. **Service Quality Classification and Tip Guidelines**:
* Based on the overall average score, classify the service quality as follows:
* Poor Service: Overall average score < 60 (Tip range: 0% ~ 5% of the bill)
* Average Service: Overall average score β‰₯ 60 and < 80 (Tip range: 10% ~ 15% of the bill)
* Good Service: Overall average score β‰₯ 80 (Tip range: 15% ~ 20% of the bill)
* Select a specific tip percentage within the appropriate range. **Exception**: If ethical violations were noted (Google Review Score is 0), the Final Tip Percentage **must be 0%**.
* Calculate the tip amount by multiplying the determined bill amount by the chosen tip percentage (round to two decimal places).
* Calculate the final total bill by adding the tip amount to the subtotal (round to two decimal places).
5. **Review Prioritization and Score Adjustment**:
* Even though all four factors are evaluated equally, if the user's review **explicitly states that specific issues previously mentioned in Google Reviews (especially ethical violations like racism) have been resolved or are no longer present**, then the Google Review score should be adjusted upwards from its potentially reduced value (e.g., from 0 back towards a score reflecting the *current* situation described by the user). The user's direct, specific, and current assessment takes precedence over older or contradicted Google Review points in such cases. Ensure the User Review Score reflects the user's sentiment.
### Ethical Violations and Tip Adjustment ###
* If there are any racist, discriminatory, or offensive remarks found in the Google reviews, regardless of the overall review sentiment or scores from other factors, the Google Review Score is automatically set to **0**. Consequently, the **Final Tip Percentage must be set to 0%** to strongly reflect the severity and unacceptability of such behavior.
### Recent Google Review ###
#### Google Review 1 ####
[4.0 stars] The atmosphere, the taste of the steak, and the service were all good. It was unique and fun to be able to choose after tasting the dessert tea and coffee. The regrettable points are: 1. The degree of meat cooking must be uniform in the course meal. Everyone has different tastes, but if the degree of cooking must be uniform because it is a course meal, I think I will consider visiting again. Those who are thinking about a course meal, please take note. 2. When serving all the course meals, they pretended not to notice that they spilled something on the table^^ and when they poured water, they spilled a lot, which was a bit embarrassing. 3. I think they could have explained it more comfortably before serving.
#### Google Review 2 ####
[4.0 stars] A steak restaurant with an American-style atmosphere. Thick tenderloin and strips served? The sound stimulated my appetite, but there was a lot of food, so I left it behind. It was a bit difficult to eat because it was undercooked in the middle. I also had stir-fried kimchi as a garnish, and it was a little sweet, so it tasted like Southeast Asia. The ice cream I had for dessert was delicious, and there was a lot of food left over.
### Video Caption Input ###
# Multiple captions can be provided under this section
Video Caption(s):
{{caption_text}}
### Output ###
Return your answer in the exact format below, providing the text analyses first, followed by the JSON block containing the final calculations:
Video Text Analysis: [Provide a summary of the **significant staff actions, expressions, and interactions** observed across all provided **Video Caption(s)**. Explain how these observations contribute to the overall service impression. Include the single, **holistically determined Video Service Score** (out of 100) based on this combined analysis.]
Recent Google Review Analysis: [Summary of the insights from the Google Reviews, including any negative or racist comments, with the assigned score (out of 100). If ethical violations result in a 0 score, state this clearly.]
User Review Analysis: [Summary of the user's review including any improvements or enhanced service mentions, with the assigned score (out of 100). Note if this review led to adjustments in the Google Review score based on explicit statements about resolved issues.]
Star Rating Analysis: [Interpret the user's star rating (e.g., converting 5/5 to 100 points) and include the assigned score (out of 100).]
Overall Analysis: [Step-by-step explanation detailing:
- How the bill amount was determined;
- How each of the four components was scored, including the holistic evaluation of video captions and any adjustments made to the Google Review score based on specific user statements about resolved issues;
- If any review mentioned racist, sexist, or other ethical violations, explicitly state that the Google Review score was set to 0 and the **Final Tip Percentage is consequently set to 0%**, reflecting the severity;
- How the overall average score was calculated;
- The reasoning for the final service quality classification based on the average score;
- How the final tip percentage was chosen (either based on the guideline range or mandatorily set to 0% due to ethical violations) and the detailed calculation for the tip amount and final total bill.
### Final Calculation Results (JSON Format) ###
```json
{{{{
"final_tip_percentage": <calculated_percentage_int>,
"final_tip_amount": <calculated_tip_float>,
"final_total_bill": <calculated_total_bill_float>
}}}}
```
### GUIDE ###
In Final Answer, The ### Final Calculation Results (JSON Format) ### section must strictly follow the JSON structure provided above, containing only the JSON object and its calculated numerical values (use floating-point numbers for all values). **DO NOT include the placeholders like <calculated_percentage_float> in the actual final output; replace them with the real calculated numbers.** Ensure no other special characters like **, $, % are used *within* the JSON structure itself, only standard JSON syntax (keys in double quotes, string values in double quotes, numbers without quotes). The text analysis sections preceding the JSON should still follow their specified format.
Complete all the text analysis and overall analysis sections first, and generate the ### Final Calculation Results (JSON Format) ### section last.
**THIS INSTRUCTION ENSURES THAT ONLY THE NECESSARY SPECIAL CHARACTERS THAT ARE PART OF THE REQUIRED OUTPUT FORMAT (E.G., standard JSON syntax) ARE USED, AND ANY OTHER SPECIAL CHARACTERS SUCH AS **, $, %, LATEX COMMANDS, OR ANY NON-STANDARD CHARACTERS SHOULD BE EXCLUDED FROM THE FINAL JSON BLOCK.**
### User Context ###
* Current Country: USA
* Currently Restaurant Name: The Golden Spoon (Assumed)
* Currently Calculated Subtotal: ${calculated_subtotal:.2f}
* Currently User Star Rating: {star_rating} / 5 (5 is the maximum)
* Currently User Review: {user_review}
"""
)
CUSTOM_CSS = """
#food-container {
display: grid;
grid-template-columns: repeat(3, 1fr);
gap: 10px;
overflow-y: auto;
height: 600px;
}
#qwen-button {
background-color: #8A2BE2 !important;
color: white !important;
border-color: #8A2BE2 !important;
}
#qwen-button:hover {
background-color: #7722CC !important;
}
#google-button {
background-color: #50C878 !important;
color: white !important;
border-color: #50C878 !important;
}
#google-button:hover {
background-color: #3CB371 !important;
}
"""
def __init__(self):
if not os.path.exists("images"):
print("κ²½κ³ : 'images' 폴더λ₯Ό 찾을 수 μ—†μŠ΅λ‹ˆλ‹€. μŒμ‹ 이미지가 ν‘œμ‹œλ˜μ§€ μ•Šμ„ 수 μžˆμŠ΅λ‹ˆλ‹€.")
for item in self.FOOD_ITEMS:
if not os.path.exists(item["image"]):
print(f"κ²½κ³ : 이미지 νŒŒμΌμ„ 찾을 수 μ—†μŠ΅λ‹ˆλ‹€ - {item['image']}")
# --- ModelClients (μ•Œλ¦¬λ°”λ°” Qwen API만 μ‚¬μš©) ---
class ModelClients:
def __init__(self, config: Config):
self.config = config
from openai import OpenAI as QwenOpenAI
self.qwen_client = QwenOpenAI(
api_key=config.QWEN_API_KEY,
base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
)
self.gemini_client = genai.Client(api_key=config.GEMINI_API_key)
def encode_video_qwen(self, video_path):
with open(video_path, "rb") as video_file:
return base64.b64encode(video_file.read()).decode("utf-8")
# --- VideoProcessor: λΉ„λ””μ˜€ ν”„λ ˆμž„ μΆ”μΆœ ---
class VideoProcessor:
def extract_video_frames(self, video_path, output_folder=None, fps=1):
if not video_path:
return [], None
if output_folder is None:
output_folder = f"frames_list/frames_{uuid.uuid4().hex}"
os.makedirs(output_folder, exist_ok=True)
cap = cv2.VideoCapture(video_path)
if not cap.isOpened():
print(f"였λ₯˜: λΉ„λ””μ˜€ νŒŒμΌμ„ μ—΄ 수 μ—†μŠ΅λ‹ˆλ‹€ - {video_path}")
return [], None
frame_paths = []
frame_rate = cap.get(cv2.CAP_PROP_FPS)
if not frame_rate or frame_rate == 0:
print("κ²½κ³ : FPSλ₯Ό 읽을 수 μ—†μŠ΅λ‹ˆλ‹€, κΈ°λ³Έκ°’ 4으둜 μ„€μ •ν•©λ‹ˆλ‹€.")
frame_rate = 4.0
frame_interval = int(frame_rate / fps) if fps > 0 else 1
if frame_interval <= 0:
frame_interval = 1
frame_count = 0
saved_frame_count = 0
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
if frame is None:
print(f"κ²½κ³ : {frame_count}번째 ν”„λ ˆμž„μ΄ λΉ„μ–΄μžˆμŠ΅λ‹ˆλ‹€.")
frame_count += 1
continue
if frame_count % frame_interval == 0:
frame_path = os.path.join(output_folder, f"frame_{saved_frame_count}.jpg")
try:
if cv2.imwrite(frame_path, frame):
frame_paths.append(frame_path)
saved_frame_count += 1
else:
print(f"κ²½κ³ : {frame_path} μ €μž₯ μ‹€νŒ¨.")
except Exception as e:
print(f"κ²½κ³ : ν”„λ ˆμž„ μ €μž₯ 였λ₯˜ ({frame_path}): {e}")
frame_count += 1
cap.release()
if not frame_paths:
print("κ²½κ³ : ν”„λ ˆμž„ μΆ”μΆœ μ‹€νŒ¨.")
if os.path.exists(output_folder):
shutil.rmtree(output_folder)
return [], None
return frame_paths, output_folder
def cleanup_temp_files(self, video_path, frame_folder):
if video_path and "temp_video_" in video_path and os.path.exists(video_path):
try:
os.remove(video_path)
print(f"μž„μ‹œ λΉ„λ””μ˜€ 파일 μ‚­μ œ: {video_path}")
except OSError as e:
print(f"μž„μ‹œ λΉ„λ””μ˜€ 파일 μ‚­μ œ 였λ₯˜: {e}")
if frame_folder and os.path.exists(frame_folder):
try:
shutil.rmtree(frame_folder)
print(f"ν”„λ ˆμž„ 폴더 μ‚­μ œ: {frame_folder}")
except OSError as e:
print(f"ν”„λ ˆμž„ 폴더 μ‚­μ œ 였λ₯˜: {e}")
# --- TipCalculator (μ•Œλ¦¬λ°”λ°” Qwen APIλ₯Ό μ‚¬μš©ν•œ 팁 계산) ---
class TipCalculator:
def __init__(self, config: Config, model_clients: ModelClients, video_processor: VideoProcessor):
self.config = config
self.model_clients = model_clients
self.video_processor = video_processor
def parse_llm_output(self, output_text: str) -> Tuple[str, float, float, float, str]:
analysis = "Analysis not found."
tip_percentage = 0.0
tip_amount = 0.0
total_bill = 0.0
# ------------------------------------------------------------------
# 1) JSON 블둝 탐색 ```json ... ``` λ˜λŠ” { "final_tip_percentage": ... }
# ------------------------------------------------------------------
json_block = None
# 1‑A) ```json ... ``` νŒ¨ν„΄
m = re.search(r"```json\s*(\{.*?\})\s*```", output_text, re.DOTALL | re.IGNORECASE)
if not m:
# 1‑B) 백틱이 μ—†λŠ” 경우, 첫 번째 { ... } 쀑에 key κ°€ 포함돼 있으면 작기
m = re.search(r"(\{[^{}]*\"final_tip_percentage\"[^{}]*\})", output_text,
re.DOTALL | re.IGNORECASE)
if m:
json_block = m.group(1)
try:
json_data = json.loads(json_block)
tip_percentage = float(json_data.get("final_tip_percentage", 0.0))
tip_amount = float(json_data.get("final_tip_amount", 0.0))
total_bill = float(json_data.get("final_total_bill", 0.0))
except (ValueError, json.JSONDecodeError) as e:
print(f"[WARN] JSON νŒŒμ‹± μ‹€νŒ¨: {e}")
# ------------------------------------------------------------------
# 2) JSON이 μ—†μœΌλ©΄ ꡬ(舊) 포맷(**Final Tip …**)으둜 νŒŒμ‹±
# ------------------------------------------------------------------
if json_block is None:
# Tip %
m = re.search(r"\*\*Final Tip Percentage\*\*:\s*([0-9]+(?:\.[0-9]+)?)%", output_text,
re.IGNORECASE)
if m:
tip_percentage = float(m.group(1))
# Tip Amount
m = re.search(r"\*\*Final Tip Amount\*\*:\s*\$?\s*([0-9]+(?:\.[0-9]+)?)", output_text,
re.IGNORECASE)
if m:
tip_amount = float(m.group(1))
# Total Bill
m = re.search(r"\*\*Final Total Bill\*\*:\s*\$?\s*([0-9]+(?:\.[0-9]+)?)", output_text,
re.IGNORECASE)
if m:
total_bill = float(m.group(1))
# ------------------------------------------------------------------
# 3) Analysis μΆ”μΆœ (JSON 블둝 μ•žλΆ€λΆ„μ„ κ·ΈλŒ€λ‘œ μ“°κ±°λ‚˜, ν‚€μ›Œλ“œλ‘œ νŒŒμ‹±)
# ------------------------------------------------------------------
if json_block:
analysis_part = output_text.split(json_block)[0] # JSON μ•ž ν…μŠ€νŠΈ
else:
analysis_part = output_text
# β€œAnalysis:” λ˜λŠ” β€œOverall Analysis:” μ΄ν›„λ§Œ 잘라 μ£Όλ©΄ 더 깔끔
m = re.search(r"(?:Overall\s+Analysis|Analysis):\s*(.*)", analysis_part,
re.DOTALL | re.IGNORECASE)
if m:
analysis = m.group(1).strip()
else:
analysis = analysis_part.strip()
return analysis, tip_percentage, tip_amount, output_text
def process_tip_qwen(self, video_file_path, star_rating, user_review, calculated_subtotal, custom_prompt=None):
if not os.path.exists(video_file_path):
return "Error: λΉ„λ””μ˜€ 파일 κ²½λ‘œκ°€ μœ νš¨ν•˜μ§€ μ•ŠμŠ΅λ‹ˆλ‹€.", 0.0, 0.0, [], None, ""
base64_video = self.model_clients.encode_video_qwen(video_file_path)
omni_caption_prompt = '''
Task 1: Describe the waiters' actions in these restaurant video frames. Please check for mistakes or negative behaviors.
Task 2: Provide a short chronological summary of the entire scene.
'''
omni_result = self.model_clients.qwen_client.chat.completions.create(
model="qwen2.5-omni-7b",
messages=[
{
"role": "system",
"content": [{"type": "text", "text": "You are a helpful assistant."}],
},
{
"role": "user",
"content": [
{"type": "video_url", "video_url": {"url": f"data:;base64,{base64_video}"}},
{"type": "text", "text": omni_caption_prompt},
],
},
],
modalities=["text"],
stream=True,
stream_options={"include_usage": True},
)
all_omni_chunks = list(omni_result)
caption_text = ""
for chunk in all_omni_chunks[:-1]:
if not chunk.choices:
continue
if chunk.choices[0].delta.content:
caption_text += chunk.choices[0].delta.content
if not caption_text.strip():
caption_text = "(No caption from Omni)"
user_review = user_review.strip() if user_review else "(No user review)"
if custom_prompt is None:
prompt = self.config.DEFAULT_PROMPT_TEMPLATE.format(
calculated_subtotal=calculated_subtotal,
star_rating=star_rating,
user_review=user_review
)
else:
try:
prompt = custom_prompt.format(
calculated_subtotal=calculated_subtotal,
star_rating=star_rating,
user_review=user_review
)
except:
prompt = self.config.DEFAULT_PROMPT_TEMPLATE.format(
calculated_subtotal=calculated_subtotal,
star_rating=star_rating,
user_review=user_review
)
final_prompt = prompt.replace("{caption_text}", caption_text)
qvq_result = self.model_clients.qwen_client.chat.completions.create(
model="qwen2.5-vl-32b-instruct",
messages=[
{"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
{"role": "user", "content": [{"type": "text", "text": final_prompt}]},
],
modalities=["text"],
stream=True,
)
all_qvq_chunks = list(qvq_result)
final_reasoning = ""
final_answer = ""
is_answering = False
for c in all_qvq_chunks[:-1]:
if not c.choices:
continue
d = c.choices[0].delta
if hasattr(d, "reasoning_content") and d.reasoning_content:
final_reasoning += d.reasoning_content
if d.content:
if not is_answering:
print("\n" + "=" * 20 + "Complete Response" + "=" * 20 + "\n")
is_answering = True
final_answer += d.content
final_text = final_reasoning + "\n" + final_answer
analysis, tip_percentage, tip_amount, output_text = self.parse_llm_output(final_text)
return analysis, tip_percentage, tip_amount, [], None, output_text
def process_tip_gemini(self, video_file_path, star_rating, user_review, calculated_subtotal, custom_prompt=None):
if not video_file_path or not os.path.exists(video_file_path):
return "λΉ„λ””μ˜€ 파일 κ²½λ‘œκ°€ μœ νš¨ν•˜μ§€ μ•ŠμŠ΅λ‹ˆλ‹€.", [], None
image_captioning_prompt = '''
Task 1: Describe the actions of any waiters or staff visible in these restaurant video. Note any specific interactions, mistakes, or positive actions.
Task 2: Provide a concise overall summary of the scene depicted in the frames, in chronological order if possible.
Task 1 Output:
Task 2 Output:
'''
video_file = self.model_clients.gemini_client.files.upload(file=video_file_path)
print(f"Uploaded file info: {video_file}")
# λ™μ˜μƒμ€ 처리 쀑(Processing) μƒνƒœμ΄λ―€λ‘œ, ACTIVE μƒνƒœκ°€ 될 λ•ŒκΉŒμ§€ λŒ€κΈ°ν•©λ‹ˆλ‹€.
while video_file.state.name == "PROCESSING":
print("λ™μ˜μƒ 처리 쀑...")
time.sleep(1) # 예: 5μ΄ˆλ§ˆλ‹€ μƒνƒœ 점검
video_file = self.model_clients.gemini_client.files.get(name=video_file.name)
# 처리 μ‹€νŒ¨ μ—¬λΆ€ 확인
if video_file.state.name == "FAILED":
raise ValueError(f"파일 처리 μ‹€νŒ¨: {video_file.state.name}")
# λΉ„λ””μ˜€λ₯Ό ν”„λ‘¬ν”„νŠΈμ— ν¬ν•¨ν•΄μ„œ μΆ”λ‘  μš”μ²­
try:
caption_summary = self.model_clients.gemini_client.models.generate_content(
model="gemini-2.0-flash-lite", # μ˜ˆμ‹œ: Gemini 2.0 Flash λͺ¨λΈ μ‚¬μš©
contents=[video_file, image_captioning_prompt]
)
caption_summary = caption_summary.text
except Exception as e:
print(f"둜컬 λͺ¨λΈ ν”„λ ˆμž„ 처리 였λ₯˜: {e}")
return f"Error in processing frames with local model: {e}", None, None
user_review = user_review.strip() if user_review and user_review.strip() else "(No user review provided)"
if custom_prompt is None:
prompt = self.config.DEFAULT_PROMPT_TEMPLATE.format(
calculated_subtotal=calculated_subtotal, star_rating=star_rating, user_review=user_review
)
else:
try:
prompt = custom_prompt.format(
calculated_subtotal=calculated_subtotal, star_rating=star_rating, user_review=user_review
)
except KeyError as e:
print(f"κ²½κ³ : μ»€μŠ€ν…€ ν”„λ‘¬ν”„νŠΈμ— ν•„μš”ν•œ ν‚€κ°€ μ—†μŠ΅λ‹ˆλ‹€: {e}. κΈ°λ³Έ ν…œν”Œλ¦Ώμ„ μ‚¬μš©ν•©λ‹ˆλ‹€.")
prompt = self.config.DEFAULT_PROMPT_TEMPLATE.format(
calculated_subtotal=calculated_subtotal, star_rating=star_rating, user_review=user_review
)
final_prompt = prompt.replace("{caption_text}", caption_summary)
messages = final_prompt
video_file = self.model_clients.gemini_client.files.upload(file=video_file_path)
print(f"Uploaded file info: {video_file}")
# λ™μ˜μƒμ€ 처리 쀑(Processing) μƒνƒœμ΄λ―€λ‘œ, ACTIVE μƒνƒœκ°€ 될 λ•ŒκΉŒμ§€ λŒ€κΈ°ν•©λ‹ˆλ‹€.
while video_file.state.name == "PROCESSING":
print("λ™μ˜μƒ 처리 쀑...")
time.sleep(1) # 예: 5μ΄ˆλ§ˆλ‹€ μƒνƒœ 점검
video_file = self.model_clients.gemini_client.files.get(name=video_file.name)
# 처리 μ‹€νŒ¨ μ—¬λΆ€ 확인
if video_file.state.name == "FAILED":
raise ValueError(f"파일 처리 μ‹€νŒ¨: {video_file.state.name}")
response = self.model_clients.gemini_client.models.generate_content(
model="gemini-2.0-flash-lite", # μ˜ˆμ‹œ: Gemini 2.0 Flash λͺ¨λΈ μ‚¬μš©
contents=[video_file, messages]
)
llm_output = response.text
analysis, tip_percentage, tip_amount, output_text = self.parse_llm_output(llm_output)
return analysis, tip_percentage, tip_amount, [], None, output_text
def calculate_manual_tip(self, tip_percent, subtotal):
tip_amount = subtotal * (tip_percent / 100)
total_bill = subtotal + tip_amount
analysis_output = f"Manual calculation using fixed tip percentage of {tip_percent}%."
tip_output = f"${tip_amount:.2f} ({tip_percent:.1f}%)"
total_bill_output = f"${total_bill:.2f}"
return analysis_output, tip_output, total_bill_output
# --- UIHandler: Gradio μΈν„°νŽ˜μ΄μŠ€ 이벀트 처리 (μ•Œλ¦¬λ°”λ°” API ν‚€ μž…λ ₯ 포함) ---
class UIHandler:
def __init__(self, config: Config, tip_calculator: TipCalculator, video_processor: VideoProcessor):
self.config = config
self.tip_calculator = tip_calculator
self.video_processor = video_processor
def update_subtotal_and_prompt(self, *args):
num_food_items = len(self.config.FOOD_ITEMS)
quantities = args[:num_food_items]
star_rating = args[num_food_items]
user_review = args[num_food_items + 1]
calculated_subtotal = 0.0
for i in range(num_food_items):
calculated_subtotal += self.config.FOOD_ITEMS[i]['price'] * quantities[i]
user_review_text = user_review.strip() if user_review and user_review.strip() else "(No user review provided)"
updated_prompt = self.config.DEFAULT_PROMPT_TEMPLATE.format(
calculated_subtotal=calculated_subtotal,
star_rating=star_rating,
user_review=user_review_text
)
updated_prompt = updated_prompt.replace("{caption_text}", "{{caption_text}}")
return calculated_subtotal, updated_prompt
def compute_tip(self, type, video_file_obj, subtotal, star_rating, user_review, custom_prompt_text):
analysis_output = "계산을 μ‹œμž‘ν•©λ‹ˆλ‹€..."
tip_percentage = 0.0
tip_output = "$0.00"
total_bill_output = f"${subtotal:.2f}"
if video_file_obj is None:
return "였λ₯˜: λΉ„λ””μ˜€ νŒŒμΌμ„ μ—…λ‘œλ“œν•΄μ£Όμ„Έμš”.", "$0.00", total_bill_output, custom_prompt_text, gr.update(value=None)
try:
temp_video_path = f"temp_video_{uuid.uuid4().hex}.mp4"
original_path = video_file_obj.name if hasattr(video_file_obj, 'name') else video_file_obj
shutil.copyfile(original_path, temp_video_path)
print(f"μž„μ‹œ λΉ„λ””μ˜€ 파일 생성: {temp_video_path}")
except Exception as e:
print(f"μž„μ‹œ λΉ„λ””μ˜€ 파일 생성 였λ₯˜: {e}")
return f"였λ₯˜: λΉ„λ””μ˜€ νŒŒμΌμ„ μ²˜λ¦¬ν•  수 μ—†μŠ΅λ‹ˆλ‹€: {e}", "$0.00", total_bill_output, custom_prompt_text, None
frame_folder = None
try:
if type == 'qwen':
analysis, tip_percentage, tip_amount, _, _, output_text = self.tip_calculator.process_tip_qwen(
temp_video_path, star_rating, user_review, subtotal, custom_prompt_text
)
else:
analysis, tip_percentage, tip_amount, _, _, output_text = self.tip_calculator.process_tip_gemini(
temp_video_path, star_rating, user_review, subtotal, custom_prompt_text
)
if "Error" in analysis:
analysis_output = analysis
tip_amount = 0.0
else:
analysis_output = f"Tip Percentage: {tip_percentage:.1f}%\n\n{output_text}"
tip_output = f"${tip_amount:.2f} ({tip_percentage:.1f}%)"
total_bill = subtotal + tip_amount
total_bill_output = f"${total_bill:.2f}"
except Exception as e:
print(f"팁 계산 쀑 였λ₯˜ λ°œμƒ (qwen): {e}")
analysis_output = f"였λ₯˜ λ°œμƒ: {e}"
tip_output = "$0.00"
total_bill_output = f"${subtotal:.2f}"
finally:
self.video_processor.cleanup_temp_files(temp_video_path, frame_folder)
return analysis_output, tip_output, total_bill_output, custom_prompt_text, gr.update(value=None)
def auto_tip_and_invoice(self, type, video_file_obj, subtotal, star_rating, review, prompt, *quantities):
analysis, tip_disp, total_bill_disp, prompt_out, vid_out = self.compute_tip(
type, video_file_obj, subtotal, star_rating, review, prompt
)
invoice = self.update_invoice_summary(*quantities, tip_disp, total_bill_disp)
return analysis, tip_disp, total_bill_disp, prompt_out, vid_out, invoice
def update_invoice_summary(self, *args):
num_items = len(self.config.FOOD_ITEMS)
quantities = args[:num_items]
if len(args) >= num_items + 2:
tip_str = args[num_items]
total_bill_str = args[num_items + 1]
else:
tip_str = "$0.00"
total_bill_str = "$0.00"
summary = ""
for i, q in enumerate(quantities):
try:
q_val = float(q)
except:
q_val = 0
if q_val > 0:
item = self.config.FOOD_ITEMS[i]
total_price = item['price'] * q_val
summary += f"{item['name']} x{int(q_val)} : ${total_price:.2f}\n"
if summary == "":
summary = "μ£Όλ¬Έν•œ 메뉴가 μ—†μŠ΅λ‹ˆλ‹€."
summary += f"\nTip: {tip_str}\nTotal Bill: {total_bill_str}"
return summary
def manual_tip_and_invoice(self, tip_percent, subtotal, *quantities):
analysis, tip_disp, total_bill_disp = self.tip_calculator.calculate_manual_tip(tip_percent, subtotal)
invoice = self.update_invoice_summary(*quantities, tip_disp, total_bill_disp)
return analysis, tip_disp, total_bill_disp, invoice
def process_payment(self, total_bill):
return f"{total_bill} κ²°μ œλ˜μ—ˆμŠ΅λ‹ˆλ‹€."
# --- App: λͺ¨λ“  μ»΄ν¬λ„ŒνŠΈ μ—°κ²° 및 Gradio μΈν„°νŽ˜μ΄μŠ€ μ‹€ν–‰ ---
class App:
def __init__(self):
self.config = Config()
self.model_clients = ModelClients(self.config)
self.video_processor = VideoProcessor()
self.tip_calculator = TipCalculator(self.config, self.model_clients, self.video_processor)
self.ui_handler = UIHandler(self.config, self.tip_calculator, self.video_processor)
self.flask_app = Flask(__name__)
def create_gradio_blocks(self):
with gr.Blocks(title="Video Tip Calculation Interface", theme=gr.themes.Soft(),
css=self.config.CUSTOM_CSS) as interface:
gr.Markdown("## Video Tip Calculation Interface (Structured)")
# --- μ»΄ν¬λ„ŒνŠΈ λ³€μˆ˜ μ„ μ–Έ (νƒ­ ꡬ쑰 μ•ˆμ—μ„œ μ •μ˜λ  κ²ƒμž„) ---
# 이 λ³€μˆ˜λ“€μ€ μ•„λž˜ νƒ­ λ‚΄λΆ€μ—μ„œ μ‹€μ œ μ»΄ν¬λ„ŒνŠΈμ— ν• λ‹Ήλ©λ‹ˆλ‹€.
quantity_inputs = []
subtotal_display = gr.Number(label="Subtotal ($)", value=0.0, interactive=False, visible=False)
subtotal_visible_display_output = None
review_input, rating_input = None, None
btn_5, btn_10, btn_15, btn_20, btn_25 = None, None, None, None, None
qwen_btn = None
gemini_btn = None
tip_display, total_bill_display, payment_btn, payment_result = None, None, None, None
video_input = None
analysis_display, order_summary_display = None, None
prompt_editor = None # ν”„λ‘¬ν”„νŠΈ μ—λ””ν„°λŠ” λ‹€λ₯Έ 탭에 μ •μ˜λ  κ²ƒμž„
# --- μ΅œμƒμœ„ λ ˆλ²¨μ— νƒ­ ꡬ성 ---
with gr.Tabs():
# --- νƒ­ 1: 메인 μΈν„°νŽ˜μ΄μŠ€ (μ›λž˜ λ ˆμ΄μ•„μ›ƒ μœ μ§€) ---
with gr.TabItem("Main Interface"):
# 이 νƒ­ μ•ˆμ— μ›λž˜μ˜ 2단 λ ˆμ΄μ•„μ›ƒμ„ κ·ΈλŒ€λ‘œ μž¬ν˜„
with gr.Row():
# --- μ™Όμͺ½ μ—΄ (μ›λž˜λŒ€λ‘œ) ---
with gr.Column(scale=2):
gr.Markdown("### 1. Select Food Items")
with gr.Column(elem_id="food-container"):
for item in self.config.FOOD_ITEMS:
with gr.Column():
gr.Image(value=item["image"], label=None, show_label=False, width=150,
height=150, interactive=False)
gr.Markdown(f"**{item['name']}** (${item['price']:.2f})")
q_input = gr.Number(label="Qty", value=0, minimum=0, step=1,
elem_id=f"qty_{item['name'].replace(' ', '_')}")
quantity_inputs.append(q_input)
gr.Markdown("### Subtotal")
subtotal_visible_display_output = gr.Textbox(value="$0.00", label="Subtotal",
interactive=False)
gr.Markdown("### 2. Service Feedback")
review_input = gr.Textbox(label="Review", placeholder="μ„œλΉ„μŠ€ 리뷰 μž‘μ„±", lines=3)
rating_input = gr.Radio(choices=[1, 2, 3, 4, 5], value=3, label="⭐Star Rating (1-5)⭐",
type="value")
gr.Markdown("### 3. Calculate Tip (Manual)")
with gr.Row():
btn_5 = gr.Button("5%")
btn_10 = gr.Button("10%")
btn_15 = gr.Button("15%")
btn_20 = gr.Button("20%")
btn_25 = gr.Button("25%")
with gr.Row():
# Qwen λ²„νŠΌ μ •μ˜λ₯Ό μ—¬κΈ°λ‘œ 이동!
gemini_btn = gr.Button("Google-Gemini AI Tip Calculation", variant="secondary", elem_id="google-button")
qwen_btn = gr.Button("Alibaba-Qwen AI Tip Calculation", variant="primary",
elem_id="qwen-button")
gr.Markdown("### 4. Results")
tip_display = gr.Textbox(label="Calculated Tip", value="$0.00", interactive=False)
total_bill_display = gr.Textbox(label="Total Bill (Subtotal + Tip)", value="$0.00",
interactive=False)
payment_btn = gr.Button("κ²°μ œν•˜κΈ°")
payment_result = gr.Textbox(label="Payment Result", value="", interactive=False)
# --- 였λ₯Έμͺ½ μ—΄ (μ›λž˜λŒ€λ‘œ, ν”„λ‘¬ν”„νŠΈ λ””μŠ€ν”Œλ ˆμ΄ μ œμ™Έ) ---
with gr.Column(scale=1):
gr.Markdown("### 5. Upload & Prompt Access")
video_input = gr.Video(label="Upload Service Video")
gr.Markdown("### 6. AI Analysis")
analysis_display = gr.Textbox(label="AI Analysis", lines=8, max_lines=12, interactive=False)
gr.Markdown("### 7. μ²­κ΅¬μ„œ")
order_summary_display = gr.Textbox(label="μ²­κ΅¬μ„œ", value="μ£Όλ¬Έ 메뉴 μ—†μŒ", lines=8, max_lines=12,
interactive=False)
# --- νƒ­ 2: ν”„λ‘¬ν”„νŠΈ νŽΈμ§‘ μ „μš© (λ„“κ²Œ!) ---
with gr.TabItem("Edit Prompt"):
gr.Markdown("### Prompt Editor")
gr.Markdown("μžλ™ μƒμ„±λœ ν”„λ‘¬ν”„νŠΈλ₯Ό μ—¬κΈ°μ„œ ν™•μΈν•˜κ³  **직접 μˆ˜μ •**ν•  수 μžˆμŠ΅λ‹ˆλ‹€. AI 뢄석 μ‹œ 여기에 μžˆλŠ” μ΅œμ’… λ‚΄μš©μ΄ μ‚¬μš©λ©λ‹ˆλ‹€.")
# 이 νƒ­μ—λŠ” ν”„λ‘¬ν”„νŠΈ νŽΈμ§‘κΈ°λ§Œ λ°°μΉ˜ν•˜μ—¬ κ°€λ‘œ λ„ˆλΉ„λ₯Ό μ΅œλŒ€ν•œ ν™œμš©
prompt_editor = gr.Code(
label="Tip Calculation Prompt (Editable)",
language="python", # ν•˜μ΄λΌμ΄νŒ…μ΄ ν•„μš” μ—†μœΌλ©΄ "text"
value="Loading prompt...",
lines=35, # μ„Έλ‘œ 높이.
)
gr.Examples(
examples=[
# μž…λ ₯ μˆœμ„œ: [Alibaba API Key, Video, Subtotal, Star Rating, Review] + [각 μŒμ‹μ˜ Qty]
["./video/sample.mp4", 0.0, 1, "He drop the tray..so bad", 0, 0, 0, 0, 0, 2, 0, 0],
["./video/sample2.mp4", 0.0, 5, "Good service!", 0, 0, 0, 0, 0, 2, 0, 0]
],
inputs=[video_input, subtotal_display, rating_input,
review_input] + quantity_inputs,
outputs=[analysis_display, tip_display, total_bill_display, video_input, order_summary_display],
label="Example: Bad Service, Good Service"
)
# --- 이벀트 ν•Έλ“€λŸ¬ μ—°κ²° ---
# μ»΄ν¬λ„ŒνŠΈλ“€μ΄ λͺ¨λ‘ μ •μ˜λœ 후에 μ—°κ²°ν•΄μ•Ό 함
# (μ‹€μ œ μ½”λ“œμ—μ„œλŠ” μ»΄ν¬λ„ŒνŠΈ None 체크 λ˜λŠ” 더 λ‚˜μ€ ꡬ쑰화 ν•„μš”)
if all([subtotal_display, subtotal_visible_display_output, review_input, rating_input, prompt_editor,
order_summary_display, video_input, analysis_display, tip_display,
total_bill_display, payment_result, qwen_btn, gemini_btn] + quantity_inputs):
# 1. μ†Œκ³„ μ—…λ°μ΄νŠΈ (μˆ¨κ²¨μ§„ Number -> λ³΄μ΄λŠ” Textbox)
subtotal_display.change(
fn=lambda x: f"${x:.2f}",
inputs=subtotal_display,
outputs=subtotal_visible_display_output
)
# 2. μž…λ ₯ λ³€κ²½ μ‹œ -> μ†Œκ³„ 계산 및 ν”„λ‘¬ν”„νŠΈ νŽΈμ§‘κΈ°('Edit Prompt' νƒ­) μ—…λ°μ΄νŠΈ
inputs_for_prompt_update = quantity_inputs + [rating_input, review_input]
outputs_for_prompt_update = [subtotal_display, prompt_editor] # μˆ¨κ²¨μ§„ μ†Œκ³„μ™€ λ‹€λ₯Έ νƒ­μ˜ ν”„λ‘¬ν”„νŠΈ 에디터
for comp in inputs_for_prompt_update:
comp.change(
fn=self.ui_handler.update_subtotal_and_prompt,
inputs=inputs_for_prompt_update,
outputs=outputs_for_prompt_update
)
# 3. μˆ˜λŸ‰ λ³€κ²½ μ‹œ -> μ²­κ΅¬μ„œ('Main Interface' νƒ­) μ—…λ°μ΄νŠΈ
for comp in quantity_inputs:
comp.change(
fn=self.ui_handler.update_invoice_summary,
inputs=quantity_inputs,
outputs=order_summary_display
)
# 4. AI 계산 λ²„νŠΌ('Main Interface' νƒ­) 클릭 μ‹œ
# μž…λ ₯: 메인 νƒ­μ˜ μ»΄ν¬λ„ŒνŠΈλ“€ + 'Edit Prompt' νƒ­μ˜ prompt_editor
# 좜λ ₯: 메인 νƒ­μ˜ μ»΄ν¬λ„ŒνŠΈλ“€ + 'Edit Prompt' νƒ­μ˜ prompt_editor (μ—…λ°μ΄νŠΈ 될 수 μžˆμœΌλ―€λ‘œ)
qwen_btn.click(
fn=lambda vid, sub, rat, rev, prom, *qty: self.ui_handler.auto_tip_and_invoice('qwen', vid, sub,
rat, rev, prom,
*qty),
inputs=[video_input, subtotal_display, rating_input, review_input, prompt_editor] + quantity_inputs,
outputs=[analysis_display, tip_display, total_bill_display, prompt_editor, video_input,
order_summary_display]
)
gemini_btn.click(
fn=lambda vid, sub, rat, rev, prom, *qty: self.ui_handler.auto_tip_and_invoice('gemini', vid, sub,
rat, rev, prom,
*qty),
inputs=[video_input, subtotal_display, rating_input, review_input, prompt_editor] + quantity_inputs,
outputs=[analysis_display, tip_display, total_bill_display, prompt_editor, video_input,
order_summary_display]
)
# 5. μˆ˜λ™ 팁 λ²„νŠΌ('Main Interface' νƒ­) 클릭 μ‹œ
manual_tip_outputs = [analysis_display, tip_display, total_bill_display, order_summary_display]
if btn_5: btn_5.click(fn=lambda sub, *qty: self.ui_handler.manual_tip_and_invoice(5, sub, *qty),
inputs=[subtotal_display] + quantity_inputs, outputs=manual_tip_outputs)
# ... (btn_10, btn_15, btn_20, btn_25 에 λŒ€ν•΄μ„œλ„ λ™μΌν•˜κ²Œ)
if btn_10: btn_10.click(fn=lambda sub, *qty: self.ui_handler.manual_tip_and_invoice(10, sub, *qty),
inputs=[subtotal_display] + quantity_inputs, outputs=manual_tip_outputs)
if btn_15: btn_15.click(fn=lambda sub, *qty: self.ui_handler.manual_tip_and_invoice(15, sub, *qty),
inputs=[subtotal_display] + quantity_inputs, outputs=manual_tip_outputs)
if btn_20: btn_20.click(fn=lambda sub, *qty: self.ui_handler.manual_tip_and_invoice(20, sub, *qty),
inputs=[subtotal_display] + quantity_inputs, outputs=manual_tip_outputs)
if btn_25: btn_25.click(fn=lambda sub, *qty: self.ui_handler.manual_tip_and_invoice(25, sub, *qty),
inputs=[subtotal_display] + quantity_inputs, outputs=manual_tip_outputs)
# 6. 결제 λ²„νŠΌ('Main Interface' νƒ­) 클릭 μ‹œ
if payment_btn: payment_btn.click(
fn=self.ui_handler.process_payment,
inputs=[total_bill_display],
outputs=[payment_result]
)
else:
print("Warning: Component initialization might be out of order for event handlers.")
return interface
def run_gradio(self):
interface = self.create_gradio_blocks()
interface.launch(share=False)
def run_flask(self):
@self.flask_app.route("/")
def index():
return "Hello Flask"
self.flask_app.run(host="0.0.0.0", port=5000, debug=True)
if __name__ == "__main__":
app = App()
app.run_gradio()