evaluate.py

import os
from utils.io import load_json, save_json, list_json_files
from metrics.object_recall import ObjectRecallMetric
from metrics.attribute_recall import AttributeRecallMetric
from metrics.object_precision import ObjectPrecisionMetric
from metrics.attribute_precision import AttributePrecisionMetric

METRICS = [
    ObjectRecallMetric(),
    AttributeRecallMetric(),
    ObjectPrecisionMetric(),
    AttributePrecisionMetric(),
]

def evaluate_all(gt_folder, pred_folder, output_path, api_key):
    results = {}
    files = list_json_files(gt_folder)
    for file in files:
        gt_data = load_json(os.path.join(gt_folder, file))
        pred_data = load_json(os.path.join(pred_folder, file))
        per_video_result = {}
        for metric in METRICS:
            score = metric.evaluate(gt_data, pred_data, api_key)
            per_video_result[metric.name] = score
        results[file] = per_video_result

    summary = {
        metric.name: sum(r[metric.name] for r in results.values()) / len(results)
        for metric in METRICS
    }
    results["summary"] = summary
    save_json(results, output_path)

if __name__ == "__main__":
    import argparse
    parser = argparse.ArgumentParser()
    parser.add_argument("--gt_folder", required=True)
    parser.add_argument("--pred_folder", required=True)
    parser.add_argument("--output", default="results/evaluation_results.json")
    parser.add_argument("--api_key", required=True)
    args = parser.parse_args()

    evaluate_all(args.gt_folder, args.pred_folder, args.output, args.api_key)