python/llavatestprep.py

import jsonlines
import random
import os
import json

# if len(sys.argv) != 2:
#     print("Usage: python script.py <image_directory>")
#     sys.exit(1)

# # Get command-line arguments
# IMG_PATH = sys.argv[1] #directory path
#print(IMG_PATH)
# Configuration
IMG_PATHS = [
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Young-Smiling-Eyeglasses/042200.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Young-Smiling-Eyeglasses/033209.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Young-Smiling-Eyeglasses/018822.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Young/023024.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Young/052001.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Young/058170.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Smiling-Eyeglasses-Young-Beard-Bangs/011362.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Smiling-Eyeglasses-Young-Beard-Bangs/046696.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Smiling-Eyeglasses-Young-Beard-Bangs/059202.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Eyeglasses-Bangs/057969.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Eyeglasses-Bangs/058132.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Eyeglasses-Bangs/022461.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Bangs/063625.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Bangs/063630.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Bangs/063654.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Eyeglasses-Bangs/015686.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Eyeglasses-Bangs/040011.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Eyeglasses-Bangs/055742.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Bangs-Eyeglasses-Young/029877.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Bangs-Eyeglasses-Young/046520.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Beard-Bangs-Eyeglasses-Young/052232.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Bangs-Smiling/005454.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Bangs-Smiling/033267.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Bangs-Smiling/058542.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Young-Eyeglasses/045280.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Young-Eyeglasses/063402.jpg",
  "/mnt/user/myang/OneDrive_1_9-6-2023/facial_attributes/images/train/Young-Eyeglasses/011625.jpg",
]

preset_prompt = "Tell me the probability of this image being real. Answer a probability score between 0 and 100."
output_file = "/fsx/ivan.laptev/Deepfake/deepfakehtml/my_code/json/llavatestquestionsdffd5.jsonl"

def create_jsonl_file(paths, prompt, output_file):
    with jsonlines.open(output_file, mode='w') as writer:
        for i, path in enumerate(paths):
            filename = path.split('/')[-1]
            question = {
                "question_id": i,
                "image": filename,
                "text": prompt,
                "category": "conversation"
            }
            writer.write(question)
            
# def create_jsonl_file_from_subdirs(root_dir, prompt, output_file):
#     with jsonlines.open(output_file, mode='w') as writer:
#         question_id = 0
#         for subdir, _, files in os.walk(root_dir):
#             # Filter out non-image files (you can customize the extensions based on your needs)
#             image_files = [f for f in files if f.lower().endswith(('.png', '.jpg', '.jpeg', '.gif', '.bmp'))]
            
#             if len(image_files) > 5:
#                 selected_images = random.sample(image_files, 5)
#             else:
#                 selected_images = image_files

#             for filename in selected_images:
#                 question = {
#                     "question_id": question_id,
#                     "image": os.path.join(subdir, filename),
#                     "text": prompt,
#                     "category": "detail"
#                 }
#                 writer.write(question)
#                 question_id += 1
                
# def create_jsonl_file_from_subdirs(root_dir, prompt, output_file):
#     questions = []
#     question_id = 0
    
#     for subdir, _, files in os.walk(root_dir):
#         # Filter out non-image files (you can customize the extensions based on your needs)
#         image_files = [f for f in files if f.lower().endswith(('.png', '.jpg', '.jpeg', '.gif', '.bmp'))]
        
#         if len(image_files) > 5:
#             selected_images = random.sample(image_files, 5)
#         else:
#             selected_images = image_files

#         for filename in selected_images:
#             question = {
#                 "question_id": question_id,
#                 "image": os.path.join(subdir, filename),
#                 "text": prompt,
#                 "category": "detail"
#             }
#             questions.append(question)
#             question_id += 1
    
#     # Write the list of questions to a JSONL file
#     with jsonlines.open(output_file, mode='w') as writer:
#         writer.write_all(questions)
        
def create_json_array_file_from_subdirs(root_dir, prompt, output_file):
    questions = []
    question_id = 0
    
    for subdir, _, files in os.walk(root_dir):
        # Filter out non-image files (you can customize the extensions based on your needs)
        image_files = [f for f in files if f.lower().endswith(('.png', '.jpg', '.jpeg', '.gif', '.bmp'))]
        
        if len(image_files) > 5:
            selected_images = random.sample(image_files, 5)
        else:
            selected_images = image_files

        for filename in selected_images:
            question = {
                "question_id": question_id,
                "image": os.path.join(subdir, filename),
                "text": prompt,
                "category": "detail"
            }
            questions.append(question)
            question_id += 1
    
    # Write the list of questions to a JSON file as an array
    with open(output_file, 'w') as f:
        json.dump(questions, f, indent=4)
        
def create_jsonl_file_from_subdirs(root_dir, prompt, output_file):
    questions = []
    question_id = 0
    
    for subdir, _, files in os.walk(root_dir):
        # Filter out non-image files (you can customize the extensions based on your needs)
        image_files = [f for f in files if f.lower().endswith(('.png', '.jpg', '.jpeg', '.gif', '.bmp'))]
        
        # if len(image_files) > 5:
        #     selected_images = random.sample(image_files, 5)
        # else:
        #     selected_images = image_files
        selected_images = image_files

        for filename in selected_images:
            question = {
                "question_id": question_id,
                "image": os.path.join(subdir, filename),
                "text": prompt,
                "category": "detail"
            }
            questions.append(question)
            question_id += 1
    
    # Write the list of questions to a JSONL file
    with jsonlines.open(output_file, mode='w') as writer:
        writer.write_all(questions)

def create_jsonl_file_from_dir(root_dir, prompt, output_file):
    questions = []
    question_id = 0
    
    for subdir, _, files in os.walk(root_dir):
        # Filter out non-image files (you can customize the extensions based on your needs)
        image_files = [f for f in files if f.lower().endswith(('.png', '.jpg', '.jpeg', '.gif', '.bmp'))]
        
        # if len(image_files) > 5:
        #     selected_images = random.sample(image_files, 5)
        # else:
        #     selected_images = image_files
        selected_images = image_files

        for filename in selected_images:
            question = {
                "question_id": question_id,
                "image": os.path.join(subdir, filename),
                "text": prompt,
                "category": "detail"
            }
            questions.append(question)
            question_id += 1
    
    # Write the list of questions to a JSONL file
    with jsonlines.open(output_file, mode='w') as writer:
        writer.write_all(questions)

def create_jsonl_file_from_dir(root_dir, prompt, output_file):
    questions = []
    question_id = 0
    all_image_files = []

    # Traverse the directory tree and collect all image files
    for subdir, _, files in os.walk(root_dir):
        image_files = [os.path.join(subdir, f) for f in files if f.lower().endswith(('.png', '.jpg', '.jpeg', '.gif', '.bmp'))]
        all_image_files.extend(image_files)

    # Select 25 random images from the collected image files
    selected_images = random.sample(all_image_files, min(1000, len(all_image_files)))
    print(len(selected_images))
    for image_path in selected_images:
        question = {
            "question_id": question_id,
            "image": image_path,
            "text": prompt,
            "category": "detail"
        }
        questions.append(question)
        question_id += 1

    # Write the list of questions to a JSONL file
    with jsonlines.open(output_file, mode='a') as writer:
        writer.write_all(questions)
  
# create_jsonl_file(IMG_PATHS, preset_prompt, output_file)    
# create_jsonl_file_from_subdirs("/mnt/user/myang/OneDrive_1_9-6-2023/facial_components/images/test/", preset_prompt, output_file)       
create_jsonl_file_from_dir("/fsx/ivan.laptev/Deepfake/dffd/ffhq/test", preset_prompt, output_file)
# create_jsonl_file_from_dir("/fsx/ivan.laptev/Deepfake/dffd/stargan/test", preset_prompt, output_file)
create_jsonl_file_from_dir("/fsx/ivan.laptev/Deepfake/dffd/stylegan_ffhq/test", preset_prompt, output_file)
# create_jsonl_file_from_dir("/fsx/ivan.laptev/Deepfake/dffdtestt/face2face", preset_prompt, output_file)