style-master / process.py
inflaton's picture
init code
1761643
Raw History Blame Contribute Delete
8.65 kB
import os
import re
from urllib.parse import urlparse
import openai
import base64
import requests
import json
from dotenv import load_dotenv
products = (
"YOUR ROLE:\n"
"You are an image tagging expert that works for an online shopping company, "
"your job is to meticulously tag every image given in a lot of detail with "
"the purpose of using those tags for improved user search and a recommender algorithm\n\n"
"UNDERSTANDING:\n"
"First, explain your understanding of what the product is, what it might be used for "
"and who it might appeal to. Use that understanding to put the rest of the analysis and tagging into context\n\n"
"TAGGING:\n"
"You need to generate 2 types of tags and output them in a JSON format in a single file. "
"So to start with, think about what kind of tags would be useful for someone searching "
"for an item to enhance the recommender algorithm.\n\n"
"1) Image descriptions: generate tags between 20 and 50 tags that describe the item in a "
"useful way. Include descriptive tags about the image, then think about the tags that would "
"be useful for someone to find this image or for a recommender algorithm to perform well.\n\n"
"2) Category tags: there are high level categories for products, you need to first put them into "
"the appropriate categories they are, and then for each category, you need to list sub-categories "
"(description given, but please come up with relevant items yourself) - there could be more than "
"one sub-category:\n"
"- How is it for [target audience]\n"
"- Types [product type]\n"
"- Style [styles]\n"
"- Purpose [how and where the item would be used]\n\n"
"Please respond in a JSON format, in the following structure:\n\n"
"{\n"
' "product": {\n'
' "title": "{{ Product Title }}",\n'
' "description": "{{ Description of the Product\'s Features and Usage }}",\n'
' "appeal": "{{ Who the product might appeal to based on demographics like age, gender, interests, etc }}"\n'
" },\n"
' "description_tags": [\n'
' "{{ Description Tag 1 }}",\n'
' "{{ Description Tag 2 }}",\n'
" // ... additional description tags as needed, between 20 to 50 tags\n"
" ],\n"
' "category_tags": {\n'
' "target_audience": ["{{ Target Audience Option 1 }}", "{{ Target Audience Option 2 }}", "..."],\n'
' "product_type": ["{{ Product Type Option 1 }}", "{{ Product Type Option 2 }}", "..."],\n'
' "style": ["{{ Style Option 1 }}", "{{ Style Option 2 }}", "..."],\n'
' "purpose": ["{{ Purpose Option 1 }}", "{{ Purpose Option 2 }}", "..."]\n'
" }\n"
"You MUST ONLY output JSON, no other text is permitted.\n"
"}"
)
# Function to encode a single image
def encode_image(image_path):
with open(image_path, "rb") as image_file:
return base64.b64encode(image_file.read()).decode("utf-8")
# Function to save response content
def save_response_content(garments_json, image_name, content):
json_string = content.replace("```json", "").replace("```", "").strip()
data = json.loads(json_string)
file_name = image_name.split("/")[-1]
data["product"]["image"] = f"http://gen-ai.local:8000/static/garments/{file_name}"
data["product"]["id"] = file_name.split(".")[-2]
# Check if file exists and has content
try:
with open(garments_json, "r+") as file:
# Read current data from file
file_data = json.load(file)
# Append new data
file_data.append(data)
# Set file's current position at offset
file.seek(0)
# Update JSON file
json.dump(file_data, file, indent=4)
except (FileNotFoundError, json.JSONDecodeError):
# Create new file with initial data
with open(garments_json, "w") as file:
json.dump([data], file, indent=4)
def process_image_file(image_path, garments_json):
# Set the OpenAI API key from environment variables
openai.api_key = os.environ["OPENAI_API_KEY"]
base64_image = encode_image(image_path)
# Prepare the image content for payload
image_content = {
"type": "image_url",
"image_url": {"url": f"data:image/jpeg;base64,{base64_image}"},
}
# The text content remains the same
text_content = {"type": "text", "text": products}
# Combine the text content with the image contents
combined_contents = [text_content] + [image_content]
headers = {
"Content-Type": "application/json",
"Authorization": f"Bearer {openai.api_key}",
}
payload = {
"model": "gpt-4-vision-preview",
"messages": [{"role": "user", "content": combined_contents}],
"max_tokens": 4000,
}
retry = 5
while retry > 0:
response = requests.post(
"https://api.openai.com/v1/chat/completions", headers=headers, json=payload
)
retry -= 1
# Check if the response was successful
if response.status_code == 200:
response_json = response.json()
try:
content = response_json["choices"][0]["message"]["content"]
# print(content)
# Save the 'content' part of the response along with the image name
save_response_content(garments_json, image_path, content)
print(f"Response content saved to '{garments_json}'.")
break
except KeyError:
print("The 'choices' key is missing in the response. Full response:")
print(response_json)
else:
print(
f"Failed to get a successful response. Status code: {response.status_code}"
)
print("Full response:")
print(response.text)
if retry > 0:
print("Retrying ...")
def merge_array(garments, products):
ootd_server_url = os.environ.get("OOTD_SERVER_URL")
for garment in garments:
image = garment["product"]["image"]
parsed_url = urlparse(garment["product"]["image"])
garment["product"]["image"] = f"{ootd_server_url}{parsed_url.path}"
id = re.split("[/.]", image)[-2]
garment["product"]["id"] = id
garment["product"]["name"] = garment["product"]["title"]
for product in products:
if product["id"] == id:
garment["product"].update(product)
break
garments.sort(key=lambda garment: int(garment["product"]["id"]))
return garments
def merge(garments_json_path, manifest_json_path):
# Check if file exists and has content
with open(manifest_json_path, "r") as manifest_file:
products = json.load(manifest_file)
with open(garments_json_path, "r+") as file:
# Read current data from file
file_data = json.load(file)
# Append new data
file_data = merge_array(file_data, products)
# Set file's current position at offset
file.seek(0)
# Update JSON file
json.dump(file_data, file, indent=4)
print(
f"Garments JSON file updated with products from '{manifest_json_path}'."
)
def garmant_id_processed(garments_json_path, id):
if os.path.exists(garments_json_path):
with open(garments_json_path, "r") as file:
garments = json.load(file)
for garment in garments:
product_id = re.split("[/.]", garment["product"]["image"])[-2]
if product_id == id:
return True
return False
if __name__ == "__main__":
load_dotenv(override=False)
garments_image_path = os.environ.get("GARMENT_IMAGES_PATH") or "../Assets/garments"
manifest_json_path = (
os.environ.get("GARMENT_MANIFEST_FILE_PATH")
or "../Assets/garments/manifest.json"
)
garments_json_path = os.environ.get("GARMENTS_FILE_PATH") or "./data/garments.json"
for root, dirs, files in os.walk(garments_image_path):
for file in files:
image_path = f"{root}/{file}"
parts = re.split("[/.]", file)
if parts[-1] == "jpg":
print(f"Processing image file {image_path}...")
id = parts[-2]
if garmant_id_processed(garments_json_path, id):
print(f"Garment with id '{id}' already processed.")
continue
process_image_file(image_path, garments_json_path)
merge(garments_json_path, manifest_json_path)