update

Author: zhbai

LICENSE

@@ -18,7 +18,6 @@ National Laboratory, U.S. Dept. of Energy nor the names of its contributors
 may be used to endorse or promote products derived from this software
 without specific prior written permission.
 
-
 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

StFT_3D.py

@@ -1,6 +1,7 @@
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
+from einops import rearrange
 from model_utils import TransformerLayer, get_2d_sincos_pos_embed
 
 
@@ -34,19 +35,19 @@ def __init__(
         self.pos_embed = nn.Parameter(
             torch.randn(1, num_patches, dim), requires_grad=False
         )
-        self.pos_embed_f = nn.Parameter(
+        self.pos_embed_fno = nn.Parameter(
             torch.randn(1, num_patches, dim), requires_grad=False
         )
         pos_embed = get_2d_sincos_pos_embed(self.pos_embed.shape[-1], grid_size)
-        pos_embed_f = get_2d_sincos_pos_embed(self.pos_embed_f.shape[-1], grid_size)
+        pos_embed_fno = get_2d_sincos_pos_embed(self.pos_embed_fno.shape[-1], grid_size)
         self.pos_embed.data.copy_(torch.from_numpy(pos_embed).float().unsqueeze(0))
-        self.pos_embed_f.data.copy_(
-            torch.from_numpy(pos_embed_f).float().unsqueeze(0)
+        self.pos_embed_fno.data.copy_(
+            torch.from_numpy(pos_embed_fno).float().unsqueeze(0)
         )
         self.encoder_layers = nn.ModuleList(
             [TransformerLayer(dim, num_heads, mlp_dim, act) for _ in range(depth)]
         )
-        self.encoder_layers_f = nn.ModuleList(
+        self.encoder_layers_fno = nn.ModuleList(
             [TransformerLayer(dim, num_heads, mlp_dim, act) for _ in range(depth)]
         )
         self.head = nn.Sequential(nn.LayerNorm(dim), nn.Linear(dim, out_dim))
@@ -63,26 +64,34 @@ def forward(self, x):
         n, l, _, ph, pw = x.shape
         x_or = x[:, :, : self.cond_time * self.freq_in_channels]
         x_added = x[:, :, (self.cond_time * self.freq_in_channels) :]
-	x_or = x_or.permute(0, 1, 3, 4, 2).view(n, l, ph, pw, self.cond_time, self.freq_in_channels)
+        x_or = rearrange(
+            x_or,
+            "n l (t v) ph pw -> n l ph pw t v",
+            t=self.cond_time,
+            v=self.freq_in_channels,
+        )
         grid_dup = x_or[:, :, :, :, :1, -2:].repeat(1, 1, 1, 1, self.layer_indx, 1)
-
-	x_added = x_added.permute(0, 1, 3, 4, 2).view(n, l, ph, pw, self.layer_indx, self.freq_in_channels - 2)
+        x_added = rearrange(
+            x_added,
+            "n l (t v) ph pw -> n l ph pw t v",
+            t=self.layer_indx,
+            v=self.freq_in_channels - 2,
+        )
         x_added = torch.cat((x_added, grid_dup), axis=-1)
         x = torch.cat((x_or, x_added), axis=-2)
         x = self.p(x)
-	v, t = x.shape[-1], x.shape[-2]
-	x = x.permute(0, 1, 5, 4, 2, 3).view(n * l, v, t, ph, pw)
+        x = rearrange(x, "n l ph pw t v -> (n l) v t ph pw")
         x_ft = torch.fft.rfftn(x, dim=[2, 3, 4])[
             :, :, :, : self.modes[0], : self.modes[1]
         ]
         x_ft_real = (x_ft.real).flatten(1)
         x_ft_imag = (x_ft.imag).flatten(1)
-	x_ft_real = x_ft_real.view(n, l, -1)
-	x_ft_imag = x_ft_imag.view(n, l, -1)
+        x_ft_real = rearrange(x_ft_real, "(n l) D -> n l D", n=n, l=l)
+        x_ft_imag = rearrange(x_ft_imag, "(n l) D -> n l D", n=n, l=l)
         x_ft_real_imag = torch.cat((x_ft_real, x_ft_imag), axis=-1)
         x = self.linear(x_ft_real_imag)
-        x = x + self.pos_embed_f
-        for layer in self.encoder_layers_f:
+        x = x + self.pos_embed_fno
+        for layer in self.encoder_layers_fno:
             x = layer(x)
         x_real, x_imag = self.q(x).split(
             self.modes[0] * self.modes[1] * self.lift_channel, dim=-1
@@ -101,19 +110,20 @@ def forward(self, x):
         )
         out_ft[:, :, :, : self.modes[0], : self.modes[1]] = x_complex
         x = torch.fft.irfftn(out_ft, s=(1, ph, pw))
-	x = x.permute(0, 3, 4, 1, 2).view(n * l, ph, pw, -1)
+        x = rearrange(x, "(n l) v t ph pw -> (n l) ph pw (v t)", n=n, l=l, t=1)
         x = self.down(x)
-	c = x.shape[-1]
-	x_f = x.permute(0, 3, 1, 2).view(n, l, c, ph, pw)
+        x_fno = rearrange(x, "(n l) ph pw c -> n l c ph pw", n=n, l=l)
         x = x_copy
         _, _, _, ph, pw = x.shape
         x = x.flatten(2)
         x = self.token_embed(x) + self.pos_embed
         for layer in self.encoder_layers:
             x = layer(x)
         x = self.head(x)
-	x = x.view(n, l, self.out_channel, ph, pw)
-        x = x + x_f
+        x = rearrange(
+            x, "n l (c ph pw) -> n l c ph pw", c=self.out_channel, ph=ph, pw=pw
+        )
+        x = x + x_fno
         return x
 
 
@@ -206,7 +216,7 @@ def __init__(
     def forward(self, x, grid):
         grid_dup = grid[None, :, :, :].repeat(x.shape[0], x.shape[1], 1, 1, 1)
         x = torch.cat((x, grid_dup), axis=2)
-	x = x.view(x.shape[0], x.shape[1] * x.shape[2], x.shape[3], x.shape[4])
+        x = rearrange(x, "B L C H W -> B (L C) H W")
         layer_outputs = []
         patches = x
         restore_params = []
@@ -240,20 +250,26 @@ def forward(self, x, grid):
                     )
 
                     patches = patches.unfold(2, p1, step_h).unfold(3, p2, step_w)
-	            n, c, h, w, ph, pw = x.shape		
-		    patches = patches.permute(0, 2, 3, 1, 4, 5).view(n, h*w, c, ph, pw)
+                    patches = rearrange(patches, "n c h w ph pw -> n (h w) c ph pw")
+
                     processed_patches = self.blocks[depth](patches)
 
-		    patches = processed_patches.permute(0, 2, 1, 3, 4).view(n, c, h, w, ph, pw)
+                    patches = rearrange(
+                        processed_patches, "n (h w) c ph pw -> n c h w ph pw", h=h, w=w
+                    )
+
                     output = F.fold(
-			torch.reshape(patches.permute(0, 1, 4, 5, 2, 3),(n, c * ph * pw, h * w)),
+                        rearrange(patches, "n c h w ph pw -> n (c ph pw) (h w)"),
                         output_size=(H_pad, W_pad),
                         kernel_size=(p1, p2),
                         stride=(step_h, step_w),
                     )
 
                     overlap_count = F.fold(
-			torch.reshape(torch.ones_like(patches).permute(0, 1, 4, 5, 2, 3),(n, c * ph * pw, h * w)),
+                        rearrange(
+                            torch.ones_like(patches),
+                            "n c h w ph pw -> n (c ph pw) (h w)",
+                        ),
                         output_size=(H_pad, W_pad),
                         kernel_size=(p1, p2),
                         stride=(step_h, step_w),