Image-To-Drawing/differentiable_rasterizer.py at main · MasonAndrewHarrison/Image-To-Drawing · 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
import torch
import torch.nn as nn
import torch.nn.functional as F
import matplotlib.pyplot as plt
from torchvision import datasets, transforms
from torchvision.datasets import ImageFolder
from torch.utils.data import DataLoader, Dataset
import filter
import kornia.contrib
from scipy.ndimage import distance_transform_edt
import matplotlib.pyplot as plt
import numpy as np

@torch.compile
def render_lines_sdf(strokes: torch.Tensor, height: int, width: int, raw_sdf: bool) -> torch.Tensor:

    device = strokes.device

    line_count = strokes.shape[0]
    canvas = torch.zeros(1, height, width).to(device)

    pixel_x = torch.linspace(0, 1, width, device=device).view(1, 1, -1)   # (1, 1, W)
    pixel_y = torch.linspace(0, 1, height, device=device).view(1, -1, 1)  # (1, H, 1)

    x1 = strokes[:-1, 0].view(-1, 1, 1)
    y1 = strokes[:-1, 1].view(-1, 1, 1)
    x2 = strokes[1:, 0].view(-1, 1, 1)
    y2 = strokes[1:, 1].view(-1, 1, 1)

    sigma = strokes[1:, 2].view(-1, 1, 1)
    radius = strokes[1:, 3].view(-1, 1, 1)

    # AB→ = B - A
    vector_ab_x = x2 - x1
    vector_ab_y = y2 - y1
    vector_ab_length_squared = vector_ab_x**2 + vector_ab_y**2 + 1e-8

    # AP→ = P - A
    vector_ap_x = (pixel_x - x1)
    vector_ap_y = (pixel_y - y1)

    # t = (AP→ · AB→) / (||AB→||²)
    t = (vector_ap_x * vector_ab_x + vector_ap_y * vector_ab_y) / vector_ab_length_squared

    # This keeps the t value in between A and B
    t = t.clamp(0, 1)

    # AT→ = A + t * AB→
    vector_at_x = x1 + t * vector_ab_x
    vector_at_y = y1 + t * vector_ab_y

    # distance = √( (P_x - AT→_x)² + (P_y - AT→_y)² )
    distance = torch.sqrt((pixel_x - vector_at_x)**2 + (pixel_y - vector_at_y)**2 + 1e-8)

    if raw_sdf:
        sdf = distance - radius
        sdf = sdf.clamp(min=0)

        canvas = sdf.min(dim=0).values
        canvas = canvas.unsqueeze(0)

    else:

        # Sign Distance Function
        sdf = distance - radius
        sdf = sdf.clamp(min=0)

        # g(sdf) = e^( -sdf² / 2σ²)
        gaussian_sdf = torch.exp(-sdf**2 / (2 * sigma**2 + 1e-8)).clamp(min=1e-6)
        gaussian_sdf = gaussian_sdf #3 * opacity
        canvas = gaussian_sdf
        #1 - canvas

        canvas = canvas.max(dim=0).values
        canvas = canvas.unsqueeze(0)

        canvas = 1 - canvas

    return canvas


def render_point_sdf(strokes: torch.Tensor, height: int, width: int, raw_sdf: bool) -> torch.Tensor:

    device = strokes.device
    canvas = torch.zeros(1, height, width).to(device)

    pixel_x = torch.linspace(0, 1, width, device=device).repeat(1, height, 1)
    pixel_y = torch.linspace(0, 1, height, device=device).repeat(1, width, 1).permute(0, 2, 1)

    x = strokes[0, 0].view(1, 1, 1).expand(1, height, width)
    y = strokes[0, 1].view(1, 1, 1).expand(1, height, width)

    sigma = strokes[0, 2].view(1, 1, 1).expand(1, height, width)
    radius = strokes[0, 3].view(1, 1, 1).expand(1, height, width)

    # distance = √( (P_x - x)² + (P_y - y)² )
    distance = torch.sqrt((pixel_x - x)**2 + (pixel_y - y)**2 + 1e-8)

    if raw_sdf:
        sdf = distance - radius
        canvas = sdf.clamp(min=0)

    else:

        # Sign Distance Function
        sdf = distance - radius
        sdf = sdf.clamp(min=0)

        # g(sdf) = e^( -sdf² / 2σ²)
        gaussian_sdf = torch.exp(-sdf**2 / (2 * sigma**2 + 1e-8)).clamp(min=1e-6)
        canvas = gaussian_sdf

        canvas = 1 - canvas

    return canvas


def render_sdf(stroke: torch.Tensor, height: int, width: int, raw_sdf: bool) -> torch.Tensor:

    line_count = stroke.shape[0]

    if line_count == 1:
        return render_point_sdf(stroke, height, width, raw_sdf)

    elif line_count > 1:
        return render_lines_sdf(stroke, height, width, raw_sdf)

    else:
        raise Exception("Stroke size is less than 1.")


def render_sdf_batched(strokes: list, height: int, width: int, raw_sdf: bool) -> torch.Tensor:

    canvas = torch.zeros((height, width), device=strokes[0].device)
    for stroke in strokes:
        partial_canvas = render_sdf(stroke, height, width, raw_sdf=False)
        partial_canvas = partial_canvas.squeeze(0)

        print(partial_canvas.shape)

        #TODO why this blank
        plt.imshow(partial_canvas.detach().cpu(), cmap="grey")
        plt.show()


    return canvas


def image_to_sdf(image: torch.Tensor, threshold: float = 0.5) -> torch.Tensor:

    binary_np = (image > threshold).float().squeeze().cpu().numpy()
    edt = distance_transform_edt(binary_np)
    sdf = torch.from_numpy(edt).float().to(image.device).unsqueeze(0)

    return sdf

def image_to_negative_sdf(image: torch.Tensor, threshold: float = 0.5) -> torch.Tensor:

    binary_np = (image < threshold).float().squeeze().cpu().numpy()
    edt = -distance_transform_edt(binary_np)
    sdf = torch.from_numpy(edt).float().to(image.device).unsqueeze(0)

    return sdf