src/v1.py

from openai import OpenAI
import base64
import requests
import re

from diffusers import DiffusionPipeline
import torch
from PIL import Image
import os
import argparse


SD_pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-0.9", torch_dtype=torch.float16, use_safetensors=True, variant="fp16")
SD_pipe.to("cuda")

RF_pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-0.9", torch_dtype=torch.float16, use_safetensors=True, variant="fp16")
RF_pipe.to("cuda")

# Function to encode the image
def encode_image(image_path):
  with open(image_path, "rb") as image_file:
    return base64.b64encode(image_file.read()).decode('utf-8')


def vision_gpt(prompt, image_url, api_key):
    client = OpenAI(api_key=api_key)
    response = client.chat.completions.create(
        model="gpt-4-vision-preview",
        messages=[
            {
              "role": "user",
              "content": [
                  {"type": "text",
                   "text": prompt},
                  {
                      "type": "image_url",
                      "image_url": {
                          "url": f"data:image/jpeg;base64,{image_url}", },
                  },
              ],
            }
        ],
        max_tokens=600,
    )
    return response.choices[0].message.content



if __name__ == "__main__":
    
    parser = argparse.ArgumentParser(description="extract differentiating attributes of the gt object class from the mistaken object class, generate synthatic images of the gt class highlighting such attributes")
    parser.add_argument('-i', "--input_path", type=str, metavar='', required=True, help="path to input image")
    parser.add_argument('-o', "--output_path", type=str, metavar='', required=True, help="path to output folder")
    parser.add_argument('-k', "--api_key", type=str, metavar='', required=True, help="valid openai api key")
    parser.add_argument('-m', "--mistaken_class", type=str, metavar='', required=True, help="model wrongly predicted this class")
    parser.add_argument('-g', "--ground_truth_class", type=str, metavar='', required=True, help="the ground truth class of the image")
    parser.add_argument('-n', "--num_generations", type=int, metavar='', required=False, default=5, help="number of generations")
    args = parser.parse_args()
    
    gt, ms = args.ground_truth_class, args.mistaken_class
    
    
    if os.path.exists(args.output_path):
        pass
    else:
        os.mkdir(args.output_path)
    
    
    base64_image = encode_image(args.input_path)
    
    prompt = """List features of the {} in this image that make it distinct from a {}? Then, write a short and 
    concise non-artistic visual diffusion prompt of a {} that includes the above features of {} (starting 
    with 'photorealistic candid portrait of') and put it inside square brackets []. Do no mention {} in 
    your prompt and ignore unrelated background scenes.""".format(gt, ms, gt, gt, ms, ms)
    
    
    print("--------------gpt prompt--------------: \n", prompt, "\n\n")
    response = vision_gpt(prompt, base64_image, args.api_key)
    print("--------------GPT response--------------: \n", response, "\n\n")
    stable_diffusion_prompt =  re.search(r'\[(.*?)\]', response).group(1)
    print("--------------stable_diffusion_prompt-------------- \n", stable_diffusion_prompt, "\n\n")
    
    
    for i in range(args.num_generations):
        generated_images = SD_pipe(prompt=stable_diffusion_prompt, num_inference_steps=75).images
        refined_image = RF_pipe(prompt=stable_diffusion_prompt, image=generated_images).images[0]
        refined_image.save(args.output_path + "{}.png".format(i), 'PNG')