added function for implicit water

experiments/6wxb/image_params_mixed_training.json

@@ -1,9 +1,9 @@
 {
     "N_PIXELS": 128,
     "PIXEL_SIZE": 2.06,
-    "SIGMA": [4.0, 15.0],
+    "SIGMA": [0.5, 5.0],
     "MODEL_FILE": "../data/protein_models/6wxb_mixed_models.npy",
-    "ROTATIONS": false,
+    "ROTATIONS": true,
     "SHIFT": true,
     "CTF": true,
     "NOISE": true,

src/cryo_sbi/wpa_simulator/cryo_em_simulator.py

@@ -10,6 +10,7 @@
 from cryo_sbi.wpa_simulator.padding import pad_image
 from cryo_sbi.wpa_simulator.shift import apply_no_shift, apply_random_shift
 from cryo_sbi.wpa_simulator.validate_image_config import check_params
+from cryo_sbi.wpa_simulator.implicit_water import add_noise_field
 
 
 class CryoEmSimulator:

src/cryo_sbi/wpa_simulator/implicit_water.py

@@ -0,0 +1,47 @@
+import torch
+
+
+def gen_noise_field(num_pixels, num_sin_func=10, max_intensity=1e-3):
+    """Generate a noise field with a given number of sinusoidal functions.
+
+    Args:
+        num_pixels (int): Number of pixels in the noise field.
+        num_sin_func (int, optional): Number of sinusoidal functions. Defaults to 10.
+        max_intensity (float, optional): Maximum intensity of the noise field. Defaults to 1e-3.
+
+    Returns:
+        torch.Tensor: Noise field.
+    """
+
+    x = torch.linspace(-100, 100, num_pixels)
+    y = torch.linspace(-100, 100, num_pixels)
+    xx, yy = torch.meshgrid(x, y)
+
+    b = 0.6 * (torch.rand((num_sin_func, 2)) - 0.5)
+    c = 2 * torch.pi * (torch.rand(num_sin_func, 2) - 0.5)
+
+    noise_field = torch.zeros_like(xx, dtype=torch.double)
+    for i in range(num_sin_func):
+        noise_field += torch.sin(b[i, 0] * xx + c[i, 0]) * torch.sin(
+            b[i, 1] * yy + c[i, 1]
+        )
+    noise_field = max_intensity * (noise_field / noise_field.max())
+    return noise_field
+
+
+def add_noise_field(image, min_intensity):
+    """Add a noise field to an image.
+
+    Args:
+        image (torch.Tensor): Image of shape (n_pixels, n_pixels) or (n_channels, n_pixels, n_pixels).
+        min_intensity (float): Minimum intensity of the image.
+
+    Returns:
+        torch.Tensor: Image with noise field.
+    """
+
+    noise_field = gen_noise_field(image.shape[0], max_intensity=1e-12)
+    idx_replace = image < min_intensity
+    image[idx_replace] = noise_field[idx_replace]
+
+    return image