Implemented Harmonic Perlin Noise

priorConditionedAnnealing/noise.py

@@ -90,10 +90,10 @@ class SDE_Noise():
 
 
 class Perlin_Noise():
-    def __init__(self, known_shape=None, scale=0.1, octaves=1):
+    def __init__(self, known_shape=None, scale=0.1, octave=1):
         self.known_shape = known_shape
         self.scale = scale
-        self.octaves = octaves
+        self.octave = octave
         self.magic = 3.141592653589  # Axis offset, should be (kinda) irrational
         # We want to genrate samples, that approx ~N(0,1)
         self.normal_factor = 14/99
@@ -102,9 +102,30 @@ class Perlin_Noise():
     def __call__(self, shape):
         self.index += 1
         noise = [self.noise([self.index*self.scale, self.magic*a]) / self.normal_factor
-                 for a in range(self.known_shape[-1])]
+                 for a in range(shape[-1])]
         return th.Tensor(noise)
 
     def reset(self):
         self.index = 0
-        self.noise = PerlinNoise(octaves=self.octaves)
+        self.noise = PerlinNoise(octaves=self.octave)
+
+
+class Harmonic_Perlin_Noise():
+    def __init__(self, known_shape=None, scale=0.1, octaves=8):
+        self.known_shape = known_shape
+        self.scale = scale
+        if type(octaves) == int:
+            octaves = [1/(i+1) for i in range(octaves)]
+        octaves = np.array(octaves)
+        self.octaves = octaves / np.linalg.norm(octaves)
+        self.reset()
+
+    def __call__(self, shape):
+        harmonics = [noise(shape)*self.octaves[i] for i, noise in enumerate(self.noises)]
+        return sum(harmonics)
+
+    def reset(self):
+        self.index = 0
+        self.noises = []
+        for octave, amplitude in enumerate(self.octaves):
+            self.noises += [Perlin_Noise(known_shape=self.known_shape, scale=self.scale, octave=(octave+1))]