【多模态处理】利用GPT逐一读取本地图片并生成描述，支持崩溃后从最新进度恢复题
代码功能：
读取本地图片文件，并使用GPT模型生成图像的元数据描述。生成的结果会保存到一个JSON文件中。代码还包含了检查点机制，以便在处理过程中程序崩溃时能够从最新的位置继续生成。
核心功能
- 读取文件并设置变量：
- 从JSON文件中读取图像路径、宽度和高度等变量。
- 根据读取的变量设置prompt，调用GPT模型。
- 调用GPT模型：
- 使用openai.ChatCompletion.create方法调用GPT模型，生成图像的元数据描述。
- 将生成的结果保存到JSON文件中。
- 保存输出到JSON：
- 每处理一张图片，就将结果追加到JSON文件中。
- 使用检查点机制：
- 每处理一张图片后，保存当前处理的位置。
- 如果处理过程中出现错误，程序可以从上次保存的位置继续处理。
- 处理本地图片文件：
- 从本地文件夹读取图片文件，并对每张图片进行处理。
最后碎碎念
提供一个模板，方便大家理解其思想，使用的时候，可以和openai最基本的代码对比着看
代码（使用中转平台url）：
使用中转平台（需要设置中转平台url）：
py
from PIL import Image
import os
import base64
import openai
import pickle
import json
# 设置API密钥和中转平台URL
API_SECRET_KEY = "your_api_secret_key"
BASE_URL = "https://api.your_base_url.com/v1"
# 图像文件夹路径
image_directory_path = 'your_image_directory_path'
# 设置要处理的图像数量
number_of_images_to_process = 50
# 输出文件路径
output_file_path = "output_results.json"
# 初始化计数器
image_counter = 0
# 读取 JSON 数据文件
data_file = 'your_data_file.json'
with open(data_file, 'r') as f:
data = json.load(f)
def encode_image(image_path):
with open(image_path, "rb") as image_file:
return base64.b64encode(image_file.read()).decode('utf-8')
def get_image_details(image_path):
"""
获取图像的详细信息，包括图像ID和尺寸。
参数:image_path (str): 图像文件的路径。
返回:tuple: 包含图像ID（文件名，不包括扩展名）和图像尺寸（宽度，高度）的元组。
示例:get_image_details('path/to/image.jpg') -> ('image', (800, 600))
"""
image_filename = os.path.basename(image_path)
image_id = os.path.splitext(image_filename)[0]
with Image.open(image_path) as img:
image_size = img.size
return image_id, image_size
def chat_completions(image_path, width, height):
base64_image = encode_image(image_path)
image_id, image_size = get_image_details(image_path)
client = OpenAI(api_key=API_SECRET_KEY, base_url=BASE_URL)
response = openai.ChatCompletion.create(
model="gpt-4",
messages=[
{"role": "system", "content": "You are an assistant that provides metadata information about images."},
{"role": "user", "content": f"Image ID: {image_id}, Width: {width}, Height: {height}"}
],
max_tokens=3000,
timeout=999,
)
return response
# 初始化结果字典
results_dict = {}
# 检查是否存在检查点文件
checkpoint_file = "checkpoint.pkl"
if os.path.exists(checkpoint_file):
with open(checkpoint_file, "rb") as f:
start_index = pickle.load(f)
else:
start_index = 0
# 处理图像文件
for i, image in enumerate(data[start_index:], start=start_index):
image_name = image['image_path']
image_file = os.path.join(image_directory_path, image_name)
image_width = image['width']
image_height = image['height']
if image_name.lower().endswith(('.png', '.jpg', '.jpeg', '.tiff', '.bmp', '.gif')):
try:
response = chat_completions(image_file, image_width, image_height)
result = {image_name: response.choices[0].message['content']}
with open(output_file_path, "a") as output_file:
output_file.write(json.dumps(result) + "\n")
except Exception as e:
print(f"Error processing image {image_name}: {e}")
continue
image_counter += 1
if image_counter >= number_of_images_to_process:
break
with open(checkpoint_file, "wb") as f:
pickle.dump(i+1, f)代码（直接使用openai的key）
py
from PIL import Image
import os
import base64
import openai
import pickle
import json
# 设置API密钥
API_SECRET_KEY = "your_api_secret_key"
# 图像文件夹路径
image_directory_path = 'your_image_directory_path'
# 设置要处理的图像数量
number_of_images_to_process = 50
# 输出文件路径
output_file_path = "output_results.json"
# 初始化计数器
image_counter = 0
# 读取 JSON 数据文件
data_file = 'your_data_file.json'
with open(data_file, 'r') as f:
data = json.load(f)
def encode_image(image_path):
with open(image_path, "rb") as image_file:
return base64.b64encode(image_file.read()).decode('utf-8')
def get_image_details(image_path):
"""
获取图像的详细信息，包括图像ID和尺寸。
参数:image_path (str): 图像文件的路径。
返回:tuple: 包含图像ID（文件名，不包括扩展名）和图像尺寸（宽度，高度）的元组。
示例:get_image_details('path/to/image.jpg') -> ('image', (800, 600))
"""
image_filename = os.path.basename(image_path)
image_id = os.path.splitext(image_filename)[0]
with Image.open(image_path) as img:
image_size = img.size
return image_id, image_size
def chat_completions(image_path, width, height):
base64_image = encode_image(image_path)
image_id, image_size = get_image_details(image_path)
openai.api_key = API_SECRET_KEY
response = openai.ChatCompletion.create(
model="gpt-4",
messages=[
{"role": "system", "content": "You are an assistant that provides metadata information about images."},
{"role": "user", "content": f"Image ID: {image_id}, Width: {width}, Height: {height}"}
],
max_tokens=3000,
timeout=999,
)
return response
# 初始化结果字典
results_dict = {}
# 检查是否存在检查点文件
checkpoint_file = "checkpoint.pkl"
if os.path.exists(checkpoint_file):
with open(checkpoint_file, "rb") as f:
start_index = pickle.load(f)
else:
start_index = 0
# 处理图像文件
for i, image in enumerate(data[start_index:], start=start_index):
image_name = image['image_path']
image_file = os.path.join(image_directory_path, image_name)
image_width = image['width']
image_height = image['height']
if image_name.lower().endswith(('.png', '.jpg', '.jpeg', '.tiff', '.bmp', '.gif')):
try:
response = chat_completions(image_file, image_width, image_height)
result = {image_name: response.choices[0].message['content']}
with open(output_file_path, "a") as output_file:
output_file.write(json.dumps(result) + "\n")
except Exception as e:
print(f"Error processing image {image_name}: {e}")
continue
image_counter += 1
if image_counter >= number_of_images_to_process:
break
with open(checkpoint_file, "wb") as f:
pickle.dump(i+1, f)