Pipeline: Perlin noise for init image

1 files changed, 57 insertions, 0 deletions

diff --git a/util/noise.py b/util/noise.py
new file mode 100644
index 0000000..38ab172
--- /dev/null
+++ b/util/noise.py
@@ -0,0 +1,57 @@
+import math
+import torch
+
+# 2D Perlin noise in PyTorch https://gist.github.com/vadimkantorov/ac1b097753f217c5c11bc2ff396e0a57
+
+
+def rand_perlin_2d(shape, res, fade=lambda t: 6*t**5 - 15*t**4 + 10*t**3, dtype=None, device=None, generator=None):
+    delta = (res[0] / shape[0], res[1] / shape[1])
+    d = (shape[0] // res[0], shape[1] // res[1])
+
+    grid = torch.stack(torch.meshgrid(
+        torch.arange(0, res[0], delta[0], dtype=dtype, device=device),
+        torch.arange(0, res[1], delta[1], dtype=dtype, device=device),
+        indexing='ij'
+    ), dim=-1) % 1
+    angles = 2*math.pi*torch.rand(res[0]+1, res[1]+1, generator=generator, dtype=dtype, device=device)
+    gradients = torch.stack((torch.cos(angles), torch.sin(angles)), dim=-1)
+
+    def tile_grads(slice1, slice2): return gradients[
+        slice1[0]:slice1[1],
+        slice2[0]:slice2[1]
+    ].repeat_interleave(d[0], 0).repeat_interleave(d[1], 1)
+
+    def dot(grad, shift): return (torch.stack((
+        grid[:shape[0], :shape[1], 0] + shift[0],
+        grid[:shape[0], :shape[1], 1] + shift[1]
+    ), dim=-1) * grad[:shape[0], :shape[1]]).sum(dim=-1)
+
+    n00 = dot(tile_grads([0, -1], [0, -1]), [0,  0])
+    n10 = dot(tile_grads([1, None], [0, -1]), [-1, 0])
+    n01 = dot(tile_grads([0, -1], [1, None]), [0, -1])
+    n11 = dot(tile_grads([1, None], [1, None]), [-1, -1])
+    t = fade(grid[:shape[0], :shape[1]])
+    return math.sqrt(2) * torch.lerp(torch.lerp(n00, n10, t[..., 0]), torch.lerp(n01, n11, t[..., 0]), t[..., 1])
+
+
+def rand_perlin_2d_octaves(shape, res, octaves=1, persistence=0.5, dtype=None, device=None, generator=None):
+    noise = torch.zeros(shape, dtype=dtype, device=device)
+    frequency = 1
+    amplitude = 1
+    for _ in range(int(octaves)):
+        noise += amplitude * rand_perlin_2d(
+            shape, (frequency*res[0], frequency*res[1]), dtype=dtype, device=device, generator=generator
+        )
+        frequency *= 2
+        amplitude *= persistence
+    return noise
+
+
+def perlin_noise(batch_size: int, width: int, height: int, res=8, octaves=1, dtype=None, device=None, generator=None):
+    return torch.stack([
+        rand_perlin_2d_octaves(
+            (width, height), (res, res), octaves, dtype=dtype, device=device, generator=generator
+        ).unsqueeze(0)
+        for _
+        in range(batch_size)
+    ])