reformat_ann.py

import json
from collections import defaultdict
import os
import re

def parse_multiple_choice_query(query_text):
    """
    Parse a multiple choice query to extract the question and choices.
    
    Example input: "What happened first? A. cut or chop or slice a vegetable, fruit, or meat; B.wash vegetable or fruit or food item"
    Returns: question, choices_list
    """
    # Split by '?' to separate question from choices
    parts = query_text.split('?', 1)
    if len(parts) != 2:
        raise ValueError(f"Invalid query format: {query_text}")
    
    question = parts[0].strip() + '?'
    choices_part = parts[1].strip()
    
    # Parse choices using regex to find A. ... B. ... pattern
    choice_pattern = r'([AB]\.\s*[^;]+?)(?=\s*[AB]\.|$)'
    matches = re.findall(choice_pattern, choices_part)
    
    if len(matches) != 2:
        # Fallback: split by '; ' and clean up
        choice_parts = choices_part.split(';')
        choices = []
        for part in choice_parts:
            part = part.strip()
            if part.startswith(('A.', 'B.')):
                choices.append(part)
        
        if len(choices) != 2:
            raise ValueError(f"Could not parse choices from: {choices_part}")
    else:
        choices = [match.strip() for match in matches]
    
    return question, choices

def reformat_annotations(input_file, output_file, video_folder, is_multiple_choice=False):
    """
    Reformat the annotation file to group queries by video_uid.
    
    Args:
        input_file: Path to the input JSON file
        output_file: Path to the output JSON file
        video_folder: Path to the video folder
        is_multiple_choice: Whether this is a multiple choice dataset (order queries)
    """
    # Load the original data
    with open(input_file, 'r') as f:
        original_data = json.load(f)
    
    # Group queries by video_uid
    video_groups = defaultdict(list)
    
    for item in original_data:
        video_uid = item['video_uid']
        
        if is_multiple_choice:
            # Parse multiple choice query
            try:
                question, choices = parse_multiple_choice_query(item['query'])
                conversation_item = {
                    'question': question,
                    'choices': choices,
                    'answer': item['answer'].replace('"', ''),
                    'question_type': item['type']
                }
            except ValueError as e:
                print(f"Error parsing query: {e}")
                print(f"Original query: {item['query']}")
                continue
        else:
            # Regular yes/no query
            conversation_item = {
                'question': item['query'],
                'answer': item['answer'],
                'question_type': item['type']
            }
        
        video_groups[video_uid].append(conversation_item)
    
    # Create the new format
    reformatted_data = []
    for video_uid, conversations in video_groups.items():
        assert not video_uid.endswith(".mp4"), f"Video UID {video_uid} should not end with .mp4"
        video_path = os.path.join(video_folder, video_uid + ".mp4")
        video_dict = {
            'video_id': video_uid,
            'video_path': video_path,
            'conversations': conversations
        }
        reformatted_data.append(video_dict)
    
    # Save the reformatted data
    with open(output_file, 'w') as f:
        json.dump(reformatted_data, f, indent=4)
    
    print(f"Reformatted {len(original_data)} queries from {len(reformatted_data)} videos")
    print(f"Output saved to: {output_file}")


def examine_dataset(input_file):
    with open(input_file, 'r') as f:
        data = json.load(f)
    for item in data:
        for conversation in item['conversations']:
            if 'choices' in conversation:
                assert len(conversation['choices']) == 2
                assert conversation['answer'] in conversation['choices'], f"Answer {conversation['answer']} is not in choices {conversation['choices']}"
            else:
                assert conversation['answer'] in ['Yes', 'No']

if __name__ == "__main__":
    # # Appearance queries (yes/no questions)
    # appearance_types = ["object_appearance", "action_appearance"]
    # for question_type in appearance_types:
    #     video_folder = f"/shared/nas2/yaox11/ruisen/Episodic_Dataset/{question_type}"
    #     input_file = f"{question_type}_queries_final.json"
    #     output_file = f"{question_type}_queries_reformatted.json"
    #     reformat_annotations(input_file, output_file, video_folder, is_multiple_choice=False)
    
    # # Order queries (multiple choice questions)
    # order_types = ["object_order", "action_order"]
    # for question_type in order_types:
    #     video_folder = f"/shared/nas2/yaox11/ruisen/Episodic_Dataset/{question_type}"
    #     input_file = f"{question_type}_queries_final.json"
    #     output_file = f"{question_type}_queries_reformatted.json"
    #     reformat_annotations(input_file, output_file, video_folder, is_multiple_choice=True)

    examine_dataset("/shared/nas2/yaox11/ruisen/Episodic_Dataset/object_appearance_queries_reformatted.json")
    examine_dataset("/shared/nas2/yaox11/ruisen/Episodic_Dataset/action_appearance_queries_reformatted.json")
    examine_dataset("/shared/nas2/yaox11/ruisen/Episodic_Dataset/object_order_queries_reformatted.json")
    examine_dataset("/shared/nas2/yaox11/ruisen/Episodic_Dataset/action_order_queries_reformatted.json")