ponder.ooo
🐍🐍🐍
example images; image save perf. improvements
+105
-102
+41
-61
abelian_sandpile.py
+41
-61
abelian_sandpile.py
···
1
1
import math
2
2
import random
3
+
import time
4
+
from pathlib import Path
3
5
4
6
import torch
5
7
6
8
from util import *
7
9
8
10
@settings
9
-
def _settings():
11
+
def _s():
10
12
scale = 2 ** 10
11
13
w = scale + 1
12
14
h = scale + 1
···
14
16
iterations = 500000
15
17
debug_modulus = 10000
16
18
17
-
out_file = "hng"
19
+
offsets = [[-n,0] for n in range(17)] + [[n,0] for n in range(17)] + [[0,17]]
20
+
drop_points = [[h//2,w//2+n] for n in range(17)]
21
+
22
+
wrap_vertical = False
23
+
wrap_horizontal = True
24
+
25
+
name = "sand"
26
+
out_file = ""
18
27
19
28
device="cuda"
20
29
ctype=torch.cdouble
21
30
dtype=torch.double
22
31
23
-
@ifmain(__name__)
32
+
@ifmain(__name__, _s)
24
33
@timed
25
-
def _main():
26
-
_ = _settings()
27
-
torch.set_default_device(_.device)
28
-
torch.set_default_dtype(_.dtype)
29
-
dim = [_.h, _.w]
34
+
def _main(settings):
35
+
globals().update(settings.get())
36
+
run_dir = time.strftime(f"%d.%m.%Y/{name}_t%H.%M.%S")
37
+
Path("out/" + run_dir).mkdir(parents=True, exist_ok=True)
38
+
with open(f"out/{run_dir}/settings.py", "w") as f:
39
+
f.write(settings.src)
40
+
torch.set_default_device(device)
41
+
torch.set_default_dtype(dtype)
42
+
dim = [h, w]
30
43
31
44
grid = torch.zeros(dim, dtype=torch.int)
32
45
33
-
#offsets = [torch.tensor(x) for x in [[0,1],[0,-1],[1,0],[-1,0]]]
34
-
#offsets = [[1,1],[1,-1],[-1,1],[-1,-1],[0,2],[0,-2],[2,0],[-2,0]]
35
-
#offsets = [[0,10],[10,0],[5,10],[10,5],[5,5],[-1,0],[0,-1]]
36
-
#offsets = [[0,5],[5,0],[0,-5],[-5,0],[1,4],[4,1],[-1,4],[4,-1],[1,-4],[-4,1],[-4,-1],[-1,-4]]
37
-
offsets = [[0,3],[3,0],[1,2],[2,1],[13,3],[3,13],[1,0],[0,1],[1,1],[0,0]]
38
-
#offsets = [x for x in [
39
-
# [1,1],
40
-
# [0,1],
41
-
# [1,0],
42
-
# [0,-5],
43
-
# [-5,-5],
44
-
# [-5,0]
45
-
# ]]
46
-
47
-
max_c = 100
46
+
max_c = 500
48
47
c_total = 0
49
-
last_c_total = 100
48
+
last_c_total = 500
50
49
51
50
n = len(offsets)
52
51
53
-
for iteration in range(_.iterations):
54
-
#spots = torch.randn(dim) > 2.5
55
-
#grid += spots
56
-
#p = [_.h // 2, _.w // 2]
57
-
p = [0,0]
58
-
#p = [random.randint(0,_.h-1), random.randint(0,_.w-1)]
52
+
for iteration in range(iterations):
59
53
last = torch.clone(grid)
60
-
grid[p[0],p[1]] = n
54
+
for p in drop_points:
55
+
#grid[p[0],p[1]] += 1
56
+
grid[p[0],p[1]] = n
57
+
# todo: make this index_put_
61
58
peaks = (grid >= n).int()
62
-
#peaks = torch.nonzero(grid >= 4)
63
59
c = 0
64
60
while torch.count_nonzero(peaks):
65
-
#while len(peaks) != 0:
66
61
c += 1
67
-
#p0 = peaks[0]
68
-
#pts = [p0 + o for o in offsets]
69
-
#pts = [p for p in pts if 0 < p[0] < _.h and 0 < p[1] < _.w]
70
-
#for p in pts:
71
-
# grid[p[0],p[1]] += 1
72
-
#grid[p0[0],p0[1]] -= 4
62
+
73
63
shift = -(n * peaks)
74
64
for offset in offsets:
75
65
p_shift = torch.roll(peaks, offset, dims=[0,1])
76
-
if offset[0] < 0:
66
+
if offset[0] < 0 and not wrap_vertical:
77
67
p_shift[offset[0]:] = 0
78
-
else:
68
+
elif not wrap_vertical:
79
69
p_shift[:offset[0]] = 0
80
-
if offset[1] < 0:
70
+
if offset[1] < 0 and not wrap_horizontal:
81
71
p_shift[:,offset[1]:] = 0
82
-
else:
72
+
elif not wrap_horizontal:
83
73
p_shift[:,:offset[1]] = 0
84
74
shift += p_shift
85
75
grid += shift
86
76
87
-
#pa = torch.roll(peaks, 1, dims=0)
88
-
#pb = torch.roll(peaks, -1, dims=0)
89
-
#pc = torch.roll(peaks, 1, dims=1)
90
-
#pd = torch.roll(peaks, -1, dims=1)
91
-
#pa[0] = 0
92
-
#pb[-1] = 0
93
-
#pc[:,0] = 0
94
-
#pd[:,-1] = 0
95
-
#shift = pa + pb + pc + pd - (4 * peaks)
96
-
#grid += shift
97
77
98
78
peaks = (grid >= n).int()
99
-
#peaks = torch.nonzero(grid >= 4)
100
79
c_total += c
101
80
if c > 20:
102
81
s = "*" if c >= max_c else ""
103
82
print(f"{iteration}: {c}{s}")
104
-
if c > (0.9 * max_c):
105
-
msave(grid / (n-1), f"{_.out_file}_{iteration}_{c}")
106
-
msave(last / (n-1), f"{_.out_file}_{iteration}_prev")
83
+
if c > (0.98 * max_c):
84
+
msave(grid / (n-1), f"{run_dir}/{out_file}_{iteration}_{c}")
85
+
msave(last / (n-1), f"{run_dir}/{out_file}_{iteration}_prev")
107
86
if c > max_c:
108
87
max_c = c
109
-
if iteration % _.debug_modulus == 0:
110
-
msave(grid / (n-1), f"{_.out_file}_{iteration}_")
88
+
if iteration % debug_modulus == 0:
89
+
msave(grid / (n-1), f"{run_dir}/{out_file}_{iteration}_")
111
90
if c_total > (last_c_total * 2):
112
-
msave(grid / (n-1), f"{_.out_file}_{iteration}_ct{c_total}")
91
+
msave(grid / (n-1), f"{run_dir}/{out_file}_{iteration}_ct{c_total}")
113
92
last_c_total = c_total
114
93
115
-
msave(grid / (n-1), _.out_file)
94
+
msave(grid / (n-1), out_file)
95
+
examples/fluid.png
examples/fluid.png
This is a binary file and will not be displayed.
examples/julia_q.png
examples/julia_q.png
This is a binary file and will not be displayed.
examples/ship.png
examples/ship.png
This is a binary file and will not be displayed.
examples/slime_a.png
examples/slime_a.png
This is a binary file and will not be displayed.
examples/slime_b.png
examples/slime_b.png
This is a binary file and will not be displayed.
examples/slime_c.png
examples/slime_c.png
This is a binary file and will not be displayed.
+12
-12
fluid.py
+12
-12
fluid.py
···
8
8
9
9
@settings
10
10
def _s():
11
-
scale = 2**8
11
+
scale = 2**11
12
12
w = scale
13
13
h = scale
14
14
15
-
max_iterations = 10000
16
-
save_every = 10
15
+
max_iterations = 100000
16
+
save_every = 100
17
17
18
18
v_diffusion_rate = 2000
19
19
m_diffusion_rate = 0.5
20
-
m_timescale = 0.1
21
-
timescale = 0.01
20
+
m_timescale = 0.01
21
+
timescale = 0.001
22
22
23
23
name = "fluid"
24
24
···
132
132
133
133
#velocities = torch.zeros([2, h, w])
134
134
upsample = torch.nn.Upsample(size=[h, w], mode='bilinear')
135
-
velocities = torch.randn([2, h//32, w//32]) * 100
135
+
velocities = torch.randn([2, h//32, w//32]) * 1 #* 100
136
136
densities = torch.ones([1, h, w]) * 0.1
137
137
138
138
cg = cgrid(h, w, 0 + 0j, 4 + 4j, dtype=torch.float, ctype=torch.cfloat)
···
165
165
border_mask[:,1:h-1,1:w-1] = 0;
166
166
167
167
for iteration in range(max_iterations):
168
-
velocities[0,3 * h//4 - 10:(3*h//4),w//2-10:w//2+10] = -100
169
-
velocities[1,3 * h//4:(3*h//4)+10,w//2-10:w//2+10] = -90
170
-
velocities[0,h//4:(h//4)+10,w//2-3:w//2+3] = 100
171
-
velocities[1,h//4:(h//4)+10,w//2-3:w//2+3] = 200
168
+
velocities[0,3 * h//4 - 10:(3*h//4),w//2-10:w//2+10] = -10 * 40
169
+
velocities[1,3 * h//4:(3*h//4)+10,w//2-10:w//2+10] = -9 * 40
170
+
velocities[0,h//4:(h//4)+10,w//2-3:w//2+3] = 10 * 40
171
+
velocities[1,h//4:(h//4)+10,w//2-3:w//2+3] = 20 * 40
172
172
velocities[0] += 0.01
173
-
densities[0,3*h//4-10:(3*h//4),w//2-9:w//2+9] += 0.1
173
+
#densities[0,3*h//4-10:(3*h//4),w//2-9:w//2+9] += 0.1
174
174
#densities[0,h//4:(h//4)+10,w//2+30:w//2+55] += 0.1
175
175
densities += 0.001
176
176
velocities[0] = diffuse(velocities[0], v_diffusion_rate, opposed_h_boundary, timescale, h, w)
···
180
180
project(velocities, h, w)
181
181
densities = diffuse(densities[0], m_diffusion_rate, continuous_boundary, m_timescale, h, w).unsqueeze(0)
182
182
densities = advect(densities, velocities, m_timescale, h, w)
183
-
densities *= 0.999
183
+
densities *= 0.99
184
184
if iteration % save_every == 0:
185
185
#res = torch.cat((velocities * 0.5 + 0.5, densities), dim=0)
186
186
#save(res, f"{run_dir}/{iteration}")
+13
-13
fractal.py
+13
-13
fractal.py
···
7
7
8
8
# Settings
9
9
10
-
w = 512 * 2
10
+
w = 2**13
11
11
h = w
12
12
13
13
#x_range = [-1.8, -1.7]
14
14
#y_range = [-0.095, 0.005]
15
15
16
16
#ship
17
-
#x_start = -1.8
18
-
#x_span = 0.1
19
-
#y_start = -0.095
20
-
#y_span = 0.1
17
+
x_start = -1.8
18
+
x_span = 0.1
19
+
y_start = -0.095
20
+
y_span = 0.1
21
21
22
-
x_start = -math.e / 2
23
-
x_span = 1.54
24
-
y_start = -2
25
-
y_span = 2
22
+
#x_start = -math.e / 2
23
+
#x_span = 1.54
24
+
#y_start = -2
25
+
#y_span = 2
26
26
27
27
alpha = 2.5
28
28
29
29
x_range = [x_start, x_start + x_span]
30
30
y_range = [y_start, y_start + y_span]
31
31
32
-
final_file = "gauss_1"
32
+
final_file = "ship_big"
33
33
34
34
device = "cuda"
35
35
ctype = torch.cfloat
···
88
88
mask = torch.abs(z) < limit
89
89
iters = torch.zeros_like(mask, dtype=torch.uint8)
90
90
for i in range(iterations):
91
-
#im = torch.abs(z.imag) * j
92
-
#z = (torch.abs(z.real) + im)**2 + c_n
93
-
z = torch.exp(-alpha * z) + c_n
91
+
im = torch.abs(z.imag) * j
92
+
z = (torch.abs(z.real) + im)**2 + c_n
93
+
#z = torch.exp(-alpha * z) + c_n
94
94
95
95
if do_jitter:
96
96
z += jitter
+2
-2
raytrace.py
+2
-2
raytrace.py
+13
-9
slime.py
+13
-9
slime.py
···
20
20
21
21
max_iterations = 1000
22
22
save_mod = 100
23
+
also_save = [10, 50, 150]
23
24
24
-
particle_count = 1000000 // 1
25
-
decay_rate = 0.9
25
+
particle_count = 10000000 // 4
26
+
decay_rate = 0.1
26
27
27
-
view_angle = tau / 8
28
-
view_distance = 1.414 * 100
28
+
view_angle = tau / 5
29
+
view_distance = 1.414 * 140
29
30
30
-
direction_count = 4
31
-
turn_amount = tau / direction_count * 1.2
32
-
move_distance = 1.414 * 3
31
+
direction_count = 12
32
+
turn_amount = tau / 6#direction_count
33
+
move_distance = 1.414 * 2
33
34
34
-
blur_size = 5
35
-
blur_sigma = 1
35
+
blur_size = 1
36
+
blur_sigma = 1.5
36
37
37
38
dtype = torch.double
38
39
ctype = torch.cdouble
···
115
116
116
117
# render
117
118
if iteration % save_mod == 0 and iteration != 0:
119
+
msave(world[0], f"{run_dir}/{iteration}")
120
+
121
+
if iteration in also_save:
118
122
msave(world[0], f"{run_dir}/{iteration}")
119
123
120
124
# decay
+23
-5
util.py
+23
-5
util.py
···
1
1
import torch
2
+
import numpy as np
2
3
from PIL import Image
4
+
from safetensors.torch import save_file as sft_save
5
+
import io
3
6
4
7
class Settings(dict):
5
8
__getattr__ = dict.__getitem__
···
55
58
cpilify(x).save(f"out/{f}.png")
56
59
57
60
# monochrome, 0 to 1
61
+
def mpilify_cpu(z):
62
+
_z = z.cpu().clamp_(0,1).mul_(255).round()
63
+
z_np = _z.unsqueeze(2).expand(-1,-1,3).type(torch.uint8).numpy()
64
+
return Image.fromarray(z_np)
65
+
58
66
def mpilify(z):
59
-
z_3d = torch.stack([z, z, z])
60
-
z_norm = z_3d.clamp(0, 1)
61
-
z_np = z_norm.detach().cpu().permute(1, 2, 0).numpy()
62
-
z_bytes = (z_np * 255).round().astype("uint8")
63
-
return Image.fromarray(z_bytes)
67
+
_z = torch.clone(z).clamp_(0,1).mul_(255).round()
68
+
z_np = _z.unsqueeze(2).expand(-1, -1, 3).type(torch.uint8).cpu().numpy()
69
+
return Image.fromarray(z_np)
70
+
71
+
def msave_cpu(x, f):
72
+
mpilify_cpu(x).save(f"out/{f}.png")
64
73
65
74
def msave(x, f):
66
75
mpilify(x).save(f"out/{f}.png")
76
+
77
+
def msave_alt(x, f):
78
+
with io.BytesIO() as buffer:
79
+
mpilify(x).save(buffer, format="png")
80
+
buffer.getvalue()
81
+
#_z = torch.clone(x).clamp_(0,1).mul_(255).round()
82
+
#z_np = _z.unsqueeze(2).expand(-1, -1, 3).type(torch.uint8)
83
+
#sft_save({"":_z.type(torch.uint8)}, f"out/{f}.mono.sft")
84
+
#torch.save(z_np, "out/{f}.pt")
67
85
68
86
# 3 channels
69
87
def pilify(z):