evaluation/calculate_scores.py

import argparse
import json
import os

# Function to initialize command-line arguments
def init_arguments():
    """
    Initialize and parse command-line arguments.

    Returns:
        argparse.Namespace: Parsed command-line arguments.
    """
    parser = argparse.ArgumentParser(description='Input JSON file path')
    parser.add_argument('--input_path', '-i', type=str, required=True,
                        help='Path to the eval result JSON file.')
    return parser.parse_args()


def format_data(gt_type, data):

    try:
        r =  gt_type(data) if data is not None else None
    except Exception as e:
        print(f"Data formatting error: {e}  data: {data}")
        r = None
    return r

# Function to process task results
def process_task_results(results, gt_type):
    """
    Process the results of a single task.

    Args:
        results (list): List of task results.
        gt_type (type): Type of the ground truth value.

    Returns:
        dict: A dictionary containing scores and wrong cases.
    """
    scores = {
        'correct': 0,
        'total': 0,
        'wrong_cases': []
    }
    for item in results:
        scores['total'] += 1
        parsed_answer = format_data(gt_type, item['parsed_answer'])
        ground_truth = format_data(gt_type, item['ground_truth'])
        if parsed_answer == ground_truth and parsed_answer is not None:
            scores['correct'] += 1
        else:
            scores['wrong_cases'].append({
                'id': item['id'],
                'question': item['question'],
                'image_path': item['image_path'],
                'parsed_answer': parsed_answer,
                'ground_truth': ground_truth
            })
    return scores


# Function to print task evaluation results
def print_task_results(task_name, scores, accuracy):
    """
    Print the evaluation results of a single task.

    Args:
        task_name (str): Name of the task.
        scores (dict): Dictionary containing scores and wrong cases.
        accuracy (float): Calculated accuracy for the task.
    """
    print(f"\n{task_name} Task:")
    print(f"Total Samples: {scores['total']}")
    print(f"Correct Count: {scores['correct']}")
    print(f"Accuracy: {accuracy:.2%}")
    print(f"Wrong Cases: {len(scores['wrong_cases'])}")


# Function to get output file path
def get_output_path(input_path):
    file_name = os.path.basename(input_path)
    eval_file_name = file_name.replace('_results.json', '_evaluation.json')
    return os.path.join(os.path.dirname(input_path), eval_file_name)


# Function to calculate all scores and save results
def calculate_scores(input_path, output_path):
    """
    Calculate scores for different tasks and save the detailed results.

    Args:
        input_path (str): Path to the input JSON file.
        output_path (str): Path to the output JSON file.
    """
    with open(input_path, 'r', encoding='utf-8') as f:
        results = json.load(f)

    tasks = {
        'counting': (results.get('counting_results', []), int),
        'relations': (results.get('relations_results', []), str),
        'combination': (results.get('combination_results', []), int)
    }

    detailed_results = {}

    print("\n=== Evaluation Results ===")
    for task_name, (task_results, gt_type) in tasks.items():
        scores = process_task_results(task_results, gt_type)
        accuracy = scores['correct'] / scores['total'] if scores['total'] > 0 else 0
        print_task_results(task_name.capitalize(), scores, accuracy)
        detailed_results[f'{task_name}_task'] = {
            'total_samples': scores['total'],
            'correct_samples': scores['correct'],
            'accuracy': accuracy,
            'wrong_cases': scores['wrong_cases']
        }

    with open(output_path, 'w', encoding='utf-8') as f:
        json.dump(detailed_results, f, indent=4, ensure_ascii=False)
    print(f"\nDetailed results have been saved to {output_path}")


# Main function
def main():
    """
    Main function to run the evaluation process.

    Example:
        python calculate_scores.py -i ./result/InternVL3-2B_results.json
    """
    args = init_arguments()
    output_path = get_output_path(args.input_path)
    calculate_scores(args.input_path, output_path)


if __name__ == "__main__":
    main()