Dissertation/data.py at main · KarthikSundar2002/Dissertation · 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
163
164
165
166
167
import json
import os
import shutil

import torch
import torchvision.transforms as transforms

from torchvision.utils import save_image
from torch.utils.data import DataLoader
from torch.autograd import Variable
from os import path

import numpy as np
import cv2
import kagglehub

from PIL import Image
from networks.LineArtGenerator.dataset import UnpairedDepthDataset
from networks.LineArtGenerator.model import Generator


def download_data(out_dir="./data/"):
    with open("./Secrets/kaggle.json", "r") as file:
        data = json.load(file)
    kagglehub.auth.set_kaggle_credentials(data["username"], data["key"])

    tempPath = kagglehub.dataset_download(
        "subinium/highresolution-anime-face-dataset-512x512"
    )
    filenames = os.listdir(tempPath)
    for file_name in filenames:
        shutil.move(os.path.join(tempPath, file_name), out_dir)
    print("Downloaded Data")


def generate_line_art(input_dir, out_dir):
    net = Generator(3,1,3)
    net.load_state_dict(torch.load("./weights/line_art/anime_style/netG_A_latest.pth", map_location='cpu'))
    net.eval()

    transforms_r = [transforms.Resize((512, 512),Image.BICUBIC), transforms.ToTensor()]
    transforms_r = transforms.Compose(transforms_r)

    test_data = UnpairedDepthDataset(root=input_dir, root2='', transforms_r=transforms_r)
    dataloader = DataLoader(test_data, batch_size=1, shuffle=False)

    full_output_dir = os.path.join(out_dir, "line_art")

    if not os.path.exists(full_output_dir):
        os.makedirs(full_output_dir)

    for i, batch in enumerate(dataloader):
        img_r  = Variable(batch['r'])
        img_depth  = Variable(batch['depth'])
        real_A = img_r
        name = batch['name'][0]
        input_image = real_A
        image = net(input_image)
        save_image(image.data, full_output_dir+'/%s_out.png' % name)

def generate_svg(line_art_dir, out_dir, device="cpu"):

    model_udf, args_udf, model_ndc, args_ndc = load_model(device)
    for img_name in os.listdir(line_art_dir):
        if ("png" in img_name or "jpg" in img_name or "bmp" in img_name):
            name, _ = os.path.splitext(img_name)
            if os.path.exists(
                os.path.join(out_dir, name + ".svg")
            ):
                continue
            if 'res' in img_name or 'keypt' in img_name or 'usm' in img_name:
                continue

            if device == 'cuda':
                    # if True:
                ww, hh = Image.open(path.join(line_art_dir, img_name)).size
                longer = ww if ww > hh else hh
                resize = True if longer > 512 else False
                if resize:
                    ratio = 512 / longer
                    print("log:\timage size (%dx%d) is too large, resize to (%dx%d)" % (
                        hh, ww, int(hh * ratio), int(ww * ratio)))
            else:
                resize = False
                ratio = 1

            img, img_np, canvas_size, _ = load_img(path.join(line_art_dir, img_name), device, args_udf.up_scale,
                                                       thin=None,
                                                       line_extractor=None,
                                                       resize=resize,
                                                       path_to_out=out_dir,
                                                       resize_to=512,
                                                       ui_mode=False)

            tensor_h, tensor_w = img.shape[2], img.shape[3]

            # predict UDF from sketch
            udf_topo_pre, usm_pre_, keypt_pre_dict, keypt_pre_np = predict_UDF(
                img, img_np, model_udf, None, name)

            linemap_pre_x, linemap_pre_y, pt_map_pre, edge_maps_pre_xy = predict_SVG(
                    udf_topo_pre, model_ndc, args_udf, args_ndc, (tensor_h, tensor_w), out_path=out_dir, refine=False, name=name, to_npz=False)

            lines_pre, _, usm_applied, usm_uncertain = refine_topology(
                    D(edge_maps_pre_xy),
                    pt_map_pre,
                    usm_pre_,
                    linemap_pre_x,
                    linemap_pre_y,
                    keypt_pre_dict,
                    down_rate=8,
                    downsample=True,
                    full_auto_mode=True)
            usm_applied = usm_applied != 0
            lines_to_svg(
                    lines_pre * 2,
                    tensor_w * 2,
                    tensor_h * 2,
                    path.join(
                        out_dir,
                        name + "_refine.svg"))
            simplify_SVG(
                    path.join(
                        out_dir,
                        name + "_refine.svg"),
                    keypt_pre_dict,
                    bezier=False,
                    rdp_simplify=True,
                    epsilon=0.4,
                    skip_len=4)
            svg_pre = path.join(
                    out_dir,
                    name + "_raw.svg")
            res_pre = svg_to_numpy(svg_pre)
            if res_pre is None:
                    res_pre = np.ones((tensor_h, tensor_w, 3)) * 255
            else:
                    res_pre = res_pre[..., np.newaxis].repeat(3, axis=-1)
            h, w = res_pre.shape[0], res_pre.shape[1]
            img_np = cv2.resize(
                img_np, (w, h), interpolation=cv2.INTER_AREA)
            keypt_pre_np = cv2.resize(
                keypt_pre_np, (w, h), interpolation=cv2.INTER_AREA)

            res_pre_keypt = blend_skeletons(res_pre, (usm_applied.astype(
                int), (usm_uncertain != 0).astype(int)), usm_mode=True)
            res_pre_keypt = add_keypt(
                res_pre_keypt, keypt_pre_dict, (tensor_h, tensor_w))
            keypt_pre_list = []
            keypt_to_color = {
                "end_point": "green",
                "sharp_turn": "red",
                "junc": "blue"}
            color_list = []
            for key in keypt_pre_dict:
                for i in range(len(keypt_pre_dict[key])):
                    keypt_pre_list.append(
                        complex(*keypt_pre_dict[key][i]))
                    color_list.append(keypt_to_color[key])
            pt_num = len(keypt_pre_list)
            assert len(keypt_pre_list) == len(color_list)
            Image.fromarray(
                res_pre_keypt.astype(
                    np.uint8)).save(
                path.join(
                    out_dir,
                    name + "_usm.png"))