VidSTG/read_annotation_multithread.py

import json
import argparse
import os
import cv2
from tqdm import tqdm
from concurrent.futures import ThreadPoolExecutor, as_completed
from read_annotation import read_json_file, process_record

    
            
def main():
    parser = argparse.ArgumentParser()
    parser.add_argument('--vidstg', type=str, default='VidSTG-Dataset/annotations/train_annotations.json')
    parser.add_argument('--vidor_anno_path_base', type=str, default='vidor/train_annotation/training/')
    parser.add_argument('--vidor_path_base', type=str, default='vidor/train/video')
    parser.add_argument('--output', type=str, default='results_multithread.json')
    args = parser.parse_args()
    # vidor_data = read_json_file('vidor/train_annotation/training/0000/2401075277.json')

    vidor_anno_path_base = args.vidor_anno_path_base
    vidor_path_base = args.vidor_path_base
    
    vidstg_data = read_json_file(args.vidstg)
    
    all_results = []
        
    # Use ThreadPoolExecutor to process records in parallel
    with ThreadPoolExecutor(max_workers=6) as executor:
        # Create a tqdm progress bar
        progress = tqdm(total=len(vidstg_data))
        
        # Submit all tasks and iterate over the results as they become available
        futures = [executor.submit(process_record, record, vidor_anno_path_base, vidor_path_base) for index, record in enumerate(vidstg_data)]
    
        for future in as_completed(futures):
            all_results.extend(future.result())
            progress.update(1)
        
    with open(args.output, 'w') as json_file:
        json.dump(all_results, json_file)
            

if __name__ == "__main__":
    main()