proto-segmentation/segmentation/preprocess_pascal.py at master · gmum/proto-segmentation · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
"""
Preprocesss PASCAL VOC 2012 dataset before training a segmentation model.
https://www.cityscapes-dataset.com/

how to run run:

python -m cityscapes.preprocess preprocess-pascal {N_JOBS}
"""
import argh

import os

from tqdm import tqdm
from PIL import Image
import numpy as np
import json
import multiprocessing

SOURCE_PATH = os.environ['SOURCE_DATA_PATH']
TARGET_PATH = os.environ['DATA_PATH']

ANNOTATIONS_DIR = os.path.join(TARGET_PATH, 'annotations')
MARGIN_IMG_DIR = os.path.join(TARGET_PATH, 'img_with_margin_0')


def process_images_in_chunks(args):
    split_key, img_ids = args
    chunk_img_ids = []

    unique_classes = set()

    for img_id in img_ids:
        chunk_img_ids.append(img_id)

        # 1. Save labels
        if split_key != 'test':
            with open(os.path.join(SOURCE_PATH, f'SegmentationClassAug/{img_id}.png'), 'rb') as f:
                img = Image.open(f).convert('RGB')

            pix = np.array(img).astype(np.uint8)
            pix = pix[:, :, 0]
            unique_classes.update(set(np.unique(pix)))
            # pix.shape = (height, width, channels)
            np.save(os.path.join(ANNOTATIONS_DIR, split_key, img_id), pix)

        # 2. Save image
        input_img_path = os.path.join(SOURCE_PATH, f'JPEGImages/{img_id}.jpg')

        with open(input_img_path, 'rb') as f:
            img = Image.open(f).convert('RGB')

        output_img_path = os.path.join(MARGIN_IMG_DIR, split_key, img_id + '.png')
        img.save(output_img_path)

        # Save image as .npy for fast loading
        pix = np.array(img).astype(np.uint8)
        # pix.shape = (height, width, channels)
        np.save(os.path.join(MARGIN_IMG_DIR, split_key, img_id), pix)

    return chunk_img_ids, unique_classes


def preprocess_pascal(n_jobs: int, chunk_size: int = 10):
    n_jobs = int(n_jobs)
    print(f"Preprocessing PASCAL VOC 2012")

    os.makedirs(ANNOTATIONS_DIR, exist_ok=True)
    os.makedirs(MARGIN_IMG_DIR, exist_ok=True)

    img_ids = {
        'train_aug': [], 'train': [], 'val': [], 'test': []
    }

    split_info_dir = os.path.join(SOURCE_PATH, 'ImageSets/SegmentationAug')

    for split_key in tqdm(['train_aug', 'train', 'val', 'test'], desc='preprocessing images'):
        split_img_ids = [img_id.strip().split('/')[-1].split('.')[0]
                         for img_id in open(os.path.join(split_info_dir, f'{split_key}.txt'), 'r')]

        os.makedirs(os.path.join(MARGIN_IMG_DIR, split_key), exist_ok=True)
        os.makedirs(os.path.join(ANNOTATIONS_DIR, split_key), exist_ok=True)

        n_chunks = int(np.ceil(len(split_img_ids) / chunk_size))
        chunk_files = np.array_split(split_img_ids, n_chunks)

        parallel_args = [(split_key, chunk) for chunk in chunk_files]

        pool = multiprocessing.Pool(n_jobs)
        prog_bar = tqdm(total=len(split_img_ids), desc=f'{split_key}')

        unique_classes = set()

        for chunk_img_ids, chunk_classes in pool.imap_unordered(process_images_in_chunks, parallel_args):
            img_ids[split_key] += chunk_img_ids
            unique_classes.update(set(chunk_classes))
            prog_bar.update(len(chunk_img_ids))

        prog_bar.close()
        pool.close()

        print(f'{split_key} unique classes:', unique_classes)

    with open(os.path.join(TARGET_PATH, 'all_images.json'), 'w') as fp:
        json.dump(img_ids, fp)


if __name__ == '__main__':
    argh.dispatch_command(preprocess_pascal)