added correct kwargs to lifting layer

Author: ThijsKuipers1995

gconv/gnn/kernels/kernel.py

@@ -181,7 +181,7 @@ def forward(self, H) -> Tensor:
         weight = self.sample_Rn(
             self.weight.repeat_interleave(H.shape[0], dim=0),
             H_product.repeat(self.out_channels, *product_dims),
-            **self.sample_H_kwargs,
+            **self.sample_Rn_kwargs,
         ).view(
             self.out_channels, num_H, self.in_channels // self.groups, *self.kernel_size
         )

gconv/gnn/modules/gconv.py

@@ -5,6 +5,8 @@
 """
 from __future__ import annotations
 
+from matplotlib import pyplot as plt
+
 from gconv.gnn.kernels import (
     GroupKernel,
     GLiftingKernel,

gconv/tests/test_equivariance.py

@@ -0,0 +1,96 @@
+import sys
+
+sys.path.append("..")
+
+
+from gconv.gnn import GLiftingConvSE3
+from gconv.geometry.groups import so3 as R
+from gconv.gnn import functional as gF
+import torch
+
+from matplotlib import pyplot as plt
+
+from torch.nn.functional import grid_sample
+
+
+def plot_activations(activations):
+    B, _, H, *_ = activations.shape
+    fig = plt.figure()
+    for i in range(B):
+        for j in range(H):
+            ax = fig.add_subplot(B, H, 1 + j + i * H)
+            ax.imshow(activations[i, 1, j, 2].detach().numpy())
+            ax.axis(False)
+    plt.show()
+
+
+def test_se3_lifting_conv():
+    torch.manual_seed(0)
+
+    batch_size = 1
+    in_channels = 2
+    out_channels = 3
+    kernel_size = 5
+    group_kernel_size = 4
+    groups = 1
+    bias = False
+
+    input = torch.zeros(batch_size, in_channels, 5, 5, 5)
+    input[:, :, 2, 2, :] = 1
+
+    grid_H = torch.Tensor(
+        [
+            [
+                [1, 0, 0],
+                [0, 1, 0],
+                [0, 0, 1],
+            ],
+            [
+                [-1, 0, 0],
+                [0, -1, 0],
+                [0, 0, 1],
+            ],
+        ]
+    )
+    from math import pi
+
+    grid_H = R.matrix_z(torch.linspace(0, 2 * pi, 5)[:-1])
+
+    grid_R3 = gF.create_grid_R3(5)
+
+    grid_R3_rotated = R.left_apply_to_R3(grid_H, grid_R3)
+    input_rotated = grid_sample(
+        input.repeat(grid_H.shape[0], 1, 1, 1, 1),
+        grid_R3_rotated,
+        mode="nearest",
+        padding_mode="zeros",
+    )
+
+    model = GLiftingConvSE3(
+        in_channels,
+        out_channels,
+        kernel_size,
+        group_kernel_size=group_kernel_size,
+        padding="same",
+        groups=groups,
+        bias=bias,
+        sampling_mode="nearest",
+        sampling_padding_mode="zeros",
+        mask=True,
+        permute_output_grid=False,
+    )
+
+    output, H = model(input_rotated, grid_H)
+
+    # plot_activations(input[:, :, None])
+    # plot_activations(input_rotated[:, :, None])
+    print(output.shape)
+    plot_activations(output)
+
+
+def main():
+    test_se3_lifting_conv()
+
+
+if __name__ == "__main__":
+    main()

gconv/tests/test_gconv_e2.py

@@ -4,7 +4,7 @@
 
 import torch
 
-from gconv.nn import GLiftingConvE2, GSeparableConvE2, GConvE2
+from gconv.gnn import GLiftingConvE2, GSeparableConvE2, GConvE2
 from gconv.geometry.groups import o2
 
 

gconv/tests/test_gconv_e3.py

@@ -5,7 +5,7 @@
 
 import torch
 
-from gconv.nn import GLiftingConvE3, GSeparableConvE3, GConvE3
+from gconv.gnn import GLiftingConvE3, GSeparableConvE3, GConvE3
 from gconv.geometry.groups import o3
 
 

gconv/tests/test_kernel_se3.py

@@ -2,7 +2,7 @@
 
 sys.path.append("..")
 
-from gconv.nn import kernels
+from gconv.gnn import kernels
 from gconv.geometry.groups import so3