ParFace/src/process_face_data.py at main · jilliam/ParFace · 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
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
import os
import glob

import cv2
from facenet_pytorch import MTCNN
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import torch
from tqdm import tqdm


DATASET_INFO = {
    "FFHQ": {
        "path": "/ds/FFHQ/ffhq_lite/",
        "ext": "*.jpg"
    },
    "celeba": {
        "path": "/ds/images/celeba/img_align_celeba/",
        "ext": "*.jpg"
    },
    "MEEI": {
        "path": "/ds/MEEI/frames/",
        "ext": "*.png"
    },
    "affectnet": {
        "path": "/ds/AffectNet/",
        "ext": "*.jpg"
    }
}


def process_data(dataset="FFHQ", resize=False, out_path=''):
    dataset_path = DATASET_INFO[dataset]["path"]
    # images = sorted(glob.glob(dataset_path + DATASET_INFO[dataset]["ext"]))
    images = [y for x in os.walk(dataset_path) for y in sorted(glob.glob(os.path.join(x[0], DATASET_INFO[dataset]["ext"])))]

    device = "cuda" if torch.cuda.is_available() else "cpu"
    mtcnn = MTCNN(device=device)
    df = pd.DataFrame(columns=["image_path", "x1", "y1", "x2", "y2", "scale"])

    for i, im_path in enumerate(tqdm(images)):
        # Read image
        im = plt.imread(im_path)

        # Resize image
        if resize:
            im = cv2.resize(im, (256, 256))

        # print(im_path)
        # Get bounding box coordinates from face detection model
        im_tensor = torch.tensor(im.copy(), dtype=torch.float, device=device)
        if dataset == "MEEI":  # or dataset == "FFHQ"  # If image type is .png?
            faces = mtcnn.detect(im_tensor * 255.)
        else:
            faces = mtcnn.detect(im_tensor)

        img_name = os.path.basename(im_path)
        dir_name = os.path.basename(os.path.dirname(im_path))
        file_name = os.path.join(dir_name, img_name)

        try:
            if faces[0] is not None:
                x1, y1, x2, y2 = faces[0][0]
                scale = faces[1][0]
                data = {"image_path": file_name, "x1": [x1], "y1": [y1], "x2": [x2], "y2": [y2], "scale": [scale]}
                df = pd.concat([df, pd.DataFrame(data)], ignore_index=True)

                if resize:
                    plt.imsave(os.path.join(out_path, f"{i + 1:06d}.jpg"), im)
            else:
                print(file_name)

        except Exception as e:
            print(f"Error occurred - {e}")
            continue

    # Store bounding box and scale as annotations
    csv_file = os.path.join(out_path, dataset + "_annotations.csv")
    df.to_csv(csv_file, index=False)
    print(f"Total images: {len(images)}. Bounding boxes: {df.shape[0]}")


def process_affectnet(dataset="affectnet", out_path=''):
    dataset_path = DATASET_INFO[dataset]["path"]
    images = [y for x in os.walk(dataset_path) for y in sorted(glob.glob(os.path.join(x[0], DATASET_INFO[dataset]["ext"])))]

    df = pd.DataFrame(columns=["image_path", "face_x", "face_y", "face_width", "face_height", "facial_landmarks",
                               "expression",  "valence", "arousal"])

    for i, im_path in enumerate(tqdm(images)):
        img_name = os.path.basename(im_path)
        dir_type = os.path.basename(os.path.dirname(im_path))
        dir_name = os.path.basename(os.path.dirname(os.path.dirname(im_path)))
        file_name = os.path.join(dir_name, dir_type, img_name)  # "train_set/images/filename"

        try:
            annot_file = os.path.splitext(img_name)[0] + "_lnd.npy"
            land_file = os.path.join(dataset_path, dir_name, "annotations", annot_file)
            lands = np.load(land_file)

            annot_file = os.path.splitext(img_name)[0] + "_exp.npy"
            exp_file = os.path.join(dataset_path, dir_name, "annotations", annot_file)
            exp = np.load(exp_file)

            annot_file = os.path.splitext(img_name)[0] + "_val.npy"
            val_file = os.path.join(dataset_path, dir_name, "annotations", annot_file)
            val = np.load(val_file)

            annot_file = os.path.splitext(img_name)[0] + "_aro.npy"
            aro_file = os.path.join(dataset_path, dir_name, "annotations", annot_file)
            aro = np.load(aro_file)

            x1, y1, x2, y2 = 0.0, 0.0, 223.0, 223.0
            data = {"image_path": file_name, "face_x": [x1], "face_y": [y1], "face_width": [x2], "face_height": [y2],
                    "facial_landmarks": [lands.tolist()], "expression": [exp.tolist()],
                    "valence": [val.tolist()], "arousal": [aro.tolist()]}
            df = pd.concat([df, pd.DataFrame(data)], ignore_index=True)

            print(file_name)

        except Exception as e:
            print(f"Error occurred - {e}")
            continue

    # Store bounding box and scale as annotations
    csv_file = os.path.join(out_path, dataset + "_annotations.csv")
    df.to_csv(csv_file, index=False)
    print(f"Total images: {len(images)}. Bounding boxes: {df.shape[0]}")


if __name__ == '__main__':
    # process_data("celeba")
    # process_data("MEEI", False, '/ds/MEEI/frames/')
    # process_data("FFHQ", False, '/ds/FFHQ/')
    # process_affectnet("affectnet", '/ds/AffectNet/')
    process_affectnet("affectnet", '/ds/AffectNet/new/')