splerge-tab-aug/infer.py at master · MaxKinny/splerge-tab-aug · 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
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
import os
import argparse

import cv2
import numpy as np
import torch

import libs.utils as utils
from libs.model import SplitModel

from termcolor import cprint

if __name__ == "__main__":
    parser = argparse.ArgumentParser()

    parser.add_argument(
        "-img",
        "--test_images_dir",
        dest="test_images_dir",
        help="Path to testing data table images (generated by prepare_data.py).",
        default="test_images",
        required=True,
    )
    parser.add_argument(
        "-m",
        "--model_weights",
        dest="model_weights",
        help="path to model weights.",
        default="model/model.pth",
        required=True,
    )
    parser.add_argument(
        "-o",
        "--output_path",
        dest="output_path",
        help="path to the output directory",
        default="outputs",
        required=True,
    )
    # parser.add_argument(
    #     "-e", "--eval", dest="eval", action="store_true", help="evaluation flag"
    # )

    configs = parser.parse_args()

    os.makedirs(configs.output_path, exist_ok=True)
    os.makedirs(os.path.join(configs.output_path, "predicted_xmls"), exist_ok=True)
    # os.makedirs(os.path.join(configs.output_path, "row_out"), exist_ok=True)
    # os.makedirs(os.path.join(configs.output_path, "col_out"), exist_ok=True)
    os.makedirs(os.path.join(configs.output_path, "images"), exist_ok=True)

    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")

    cprint("creating split model...", "blue", attrs=["bold"])
    model = SplitModel(eval_mode=True).to(device)

    cprint("loading weights...", "blue", attrs=["bold"])
    model.load_state_dict(
        torch.load(configs.model_weights, map_location=device)["model_state_dict"]
    )

    model.eval()

    images = os.listdir(configs.test_images_dir)

    cprint("Predicting table rows and columns:", "green", attrs=["bold"])
    print(40 * "-")

    with torch.no_grad():
        for i, image_name in enumerate(images):
            print("[" + str(i + 1) + "/" + str(len(images)) + "]", image_name)
            image_path = os.path.join(configs.test_images_dir, image_name)
            xml_path = os.path.join(
                configs.output_path, "predicted_xmls", image_name.split(".")[0] + ".xml"
            )
            image = cv2.imread(image_path).astype("float32")
            H, W, C = image.shape
            resized_image = utils.resize_image(image).transpose((2, 0, 1))
            input_image = utils.normalize_numpy_image(resized_image).unsqueeze(0)

            rpn_out, cpn_out = model(input_image.to(device))

            rpn_image = utils.probs_to_image(
                rpn_out.detach().clone(), input_image.shape, 1
            ).cpu()
            cpn_image = utils.probs_to_image(
                cpn_out.detach().clone(), input_image.shape, 0
            ).cpu()

            grid_img, row_image, col_image = utils.binary_grid_from_prob_images(
                rpn_image, cpn_image
            )

            grid_np_img = utils.tensor_to_numpy_image(grid_img)
            row_np_image = utils.tensor_to_numpy_image(row_image)
            col_np_image = utils.tensor_to_numpy_image(col_image)

            utils.process_output(image, row_np_image, col_np_image, xml_path)

            # if configs.eval:
            #     if not os.path.exists(os.path.join(configs.output_path, "row_out")):
            #         os.mkdir(os.path.join(configs.output_path, "row_out"))
            #     if not os.path.exists(os.path.join(configs.output_path, "col_out")):
            #         os.mkdir(os.path.join(configs.output_path, "col_out"))

            #     utils.tensor_to_numpy_image(
            #         row_image,
            #         write_path=os.path.join(configs.output_path, "row_out", image_name),
            #     )
            #     utils.tensor_to_numpy_image(
            #         col_image,
            #         write_path=os.path.join(configs.output_path, "col_out", image_name),
            #     )

            grid_np_img = cv2.resize(grid_np_img, (W, H))
            grid_np_img = cv2.cvtColor(grid_np_img, cv2.COLOR_GRAY2BGR)
            test_image = image.copy()
            test_image[np.where((grid_np_img == [255, 255, 255]).all(axis=2))] = [
                0,
                255,
                0,
            ]
            cv2.imwrite(
                os.path.join(configs.output_path, "images", image_name[:-4] + ".png"),
                test_image,
            )

            row_img = image.copy()
            rpn_image[rpn_image > 0.7] = 255
            rpn_image[rpn_image <= 0.7] = 0
            rpn_image = rpn_image.squeeze(0).squeeze(0).detach().numpy()
            rpn_image = cv2.resize(rpn_image, (W, H), interpolation=cv2.INTER_NEAREST)
            rpn_image = cv2.cvtColor(rpn_image, cv2.COLOR_GRAY2BGR)
            row_img[np.where((rpn_image == [255, 255, 255]).all(axis=2))] = [
                255,
                0,
                255,
            ]
            cv2.imwrite(
                os.path.join(
                    configs.output_path, "images", image_name[:-4] + "_row.png"
                ),
                row_img,
            )

            col_img = image.copy()
            cpn_image[cpn_image > 0.7] = 255
            cpn_image[cpn_image <= 0.7] = 0
            cpn_image = cpn_image.squeeze(0).squeeze(0).detach().numpy()
            cpn_image = cv2.resize(cpn_image, (W, H), interpolation=cv2.INTER_NEAREST)
            cpn_image = cv2.cvtColor(cpn_image, cv2.COLOR_GRAY2BGR)
            col_img[np.where((cpn_image == [255, 255, 255]).all(axis=2))] = [
                255,
                0,
                255,
            ]
            cv2.imwrite(
                os.path.join(
                    configs.output_path, "images", image_name[:-4] + "_col.png"
                ),
                col_img,
            )