extract_yolo_seg_lane_marking_dataset.py

import os
import json
import random
import argparse
import shutil
from tqdm import tqdm
import yaml
import utils
from safe_executor import SafeExecutor

class_mapping = {
    "lm_dashed": 1,
    "lm_solid": 0,
    "lm_botts_dot": 0,  # Treating as lm_solid
    "lm_shaded": 0  # Treating as lm_solid
}

def extract_base_dataset(from_res):
    os.system(f"python extract_base_dataset.py --from_res {from_res}")

def remove_cache_dir(cache_dir):
    if os.path.exists(cache_dir):
        shutil.rmtree(cache_dir)

def create_cache_dir(cache_dir):
    utils.check_and_create_dir(cache_dir)

def load_annotations(file):
    with open(file) as f:
        return json.load(f)

def convert_and_save_annotations(annotated_files, cache_dir, from_res):
    width, height = map(int, from_res.split('x'))
    for file in tqdm(annotated_files, desc="Converting and saving annotations"):
        base_name = os.path.basename(file)
        output_file_path = os.path.join(cache_dir, f'{base_name}.txt')
        
        lane_annotations_path = os.path.join(file, "annotations", "lane_markings.json")
        
        try:
            lane_annotations = load_annotations(lane_annotations_path)
        except FileNotFoundError:
            with open(output_file_path, 'w') as f:
                f.write("")   
            continue 
        
        yolo_annotations = utils.convert_lane_annotations_to_yolo_seg_format(lane_annotations, class_mapping, width, height)
        
        with open(output_file_path, 'w') as f:
            if yolo_annotations:
                for line in yolo_annotations:
                    f.write(f"{line}\n")
            else:
                # Create empty file if no annotations
                f.write("")

def split_files(list_of_files, train_split=0.8):
    random.shuffle(list_of_files)
    split_index = int(len(list_of_files) * train_split)
    return list_of_files[:split_index], list_of_files[split_index:]

def prepare_yolo_dataset(train_files, val_files, from_res):
    dataset_dir = os.path.join(utils.ROOT_DIR, "dataset", f"yolo_seg_lane_{from_res}")
    train_dir = os.path.join(dataset_dir, "train")
    val_dir = os.path.join(dataset_dir, "val")
    
    if os.path.exists(dataset_dir):
        user_input = input(f"The dataset directory {dataset_dir} already exists. Do you want to remove it? (y/n): ")
        if user_input.lower() == 'y':
            shutil.rmtree(dataset_dir)
        else:
            print("Exiting without making changes.")
            return
    
    utils.check_and_create_dir(train_dir)
    utils.check_and_create_dir(val_dir)

    for file in tqdm(train_files, desc="Preparing YOLO train dataset"):
        base_name = os.path.splitext(os.path.basename(file))[0]
        image_file = os.path.join(utils.ROOT_DIR, "dataset", f'{from_res}_images', f'{base_name}.jpg')
        if os.path.exists(image_file):
            shutil.copy(os.path.join(utils.ROOT_DIR, '.cache', f'{from_res}_annotations', file), train_dir)
            shutil.copy(image_file, train_dir)

    for file in tqdm(val_files, desc="Preparing YOLO val dataset"):
        base_name = os.path.splitext(os.path.basename(file))[0]
        image_file = os.path.join(utils.ROOT_DIR, "dataset", f'{from_res}_images', f'{base_name}.jpg')
        if os.path.exists(image_file):
            shutil.copy(os.path.join(utils.ROOT_DIR, '.cache', f'{from_res}_annotations', file), val_dir)
            shutil.copy(image_file, val_dir)

    create_yaml_file(dataset_dir, train_dir, val_dir)

def create_yaml_file(dataset_dir, train_dir, val_dir):
    yaml_content = {
        'path': dataset_dir,
        'train': 'train',  # relative to 'path'
        'val': 'val',      # relative to 'path'
        'names': {
            0: 'lm_solid',
            1: 'lm_dashed',
        }
    }

    yaml_file_path = os.path.join(dataset_dir, 'dataset.yaml')
    with open(yaml_file_path, 'w') as yaml_file:
        yaml.dump(yaml_content, yaml_file, default_flow_style=False)

def main():
    parser = argparse.ArgumentParser()
    supported_resolutions = utils.get_supported_resolutions()
    str_supported_resolutions = ', '.join(supported_resolutions)
    parser.add_argument('--from_res', type=str, help=f'Choose available dataset: {str_supported_resolutions}', required=True)
    parser.add_argument('--cache_enabled', type=bool, help='Enable caching', default=False)
    args = parser.parse_args()

    if args.from_res not in supported_resolutions:
        print(f"Unsupported resolution. Supported resolutions are: {str_supported_resolutions}")
        exit(1)

    extract_base_dataset(args.from_res)

    annotated_files = utils.get_annotated_files_list()

    cache_dir = os.path.join(utils.ROOT_DIR, ".cache", f"{args.from_res}_annotations")
    if not args.cache_enabled:
        remove_cache_dir(cache_dir)
    create_cache_dir(cache_dir)

    paths_to_cleanup = [cache_dir, os.path.join(utils.ROOT_DIR, "dataset", f"yolo_seg_lane_{args.from_res}")]

    with SafeExecutor(paths_to_cleanup):
        convert_and_save_annotations(annotated_files, cache_dir, args.from_res)

        list_of_files = os.listdir(cache_dir)
        train_files, val_files = split_files(list_of_files)

        prepare_yolo_dataset(train_files, val_files, args.from_res)

    print("Annotations extracted and YOLO dataset prepared successfully")

if __name__ == "__main__":
    main()