🗃️ Increase input channels from 12 to 13

The datacube has 13 channels, namely 10 from Sentinel-2's 10m and 20m resolution bands, 2 from Sentinel-1's VV and VH, and 1 from the Copernicus DEM.

src/datamodule.py

@@ -9,7 +9,7 @@
 # %%
 class RandomDataset(torch.utils.data.Dataset):
     """
-    Torch Dataset that returns tensors of size (12, 256, 256) with random
+    Torch Dataset that returns tensors of size (13, 256, 256) with random
     values.
     """
 
@@ -20,7 +20,7 @@ def __len__(self):
         return 2048
 
     def __getitem__(self, idx: int):
-        return torch.randn(12, 256, 256)
+        return torch.randn(13, 256, 256)
 
 
 class BaseDataModule(L.LightningDataModule):

src/model.py

@@ -58,7 +58,7 @@ def __init__(self, lr: float = 0.001, mask_ratio: float = 0.75):
             layer_norm_eps=1e-12,
             image_size=256,  # default was 224
             patch_size=32,  # default was 16
-            num_channels=12,  # default was 3
+            num_channels=13,  # default was 3
             qkv_bias=True,
             decoder_num_attention_heads=16,
             decoder_hidden_size=512,
@@ -90,7 +90,7 @@ def training_step(self, batch: torch.Tensor, batch_idx: int) -> torch.Tensor:
         - https://github.com/huggingface/transformers/blob/v4.35.2/src/transformers/models/vit_mae/modeling_vit_mae.py#L948-L1010
         """
         x: torch.Tensor = batch
-        # x: torch.Tensor = torch.randn(32, 12, 256, 256)  # BCHW
+        # x: torch.Tensor = torch.randn(32, 13, 256, 256)  # BCHW
 
         # Forward encoder
         outputs_encoder: dict = self(x)
@@ -101,7 +101,7 @@ def training_step(self, batch: torch.Tensor, batch_idx: int) -> torch.Tensor:
             ids_restore=outputs_encoder.ids_restore,
         )
         # output shape (batch_size, num_patches, patch_size*patch_size*num_channels)
-        assert outputs_decoder.logits.shape == torch.Size([32, 64, 12288])
+        assert outputs_decoder.logits.shape == torch.Size([32, 64, 13312])
 
         # Log training loss and metrics
         loss: torch.Tensor = self.vit.forward_loss(
@@ -113,7 +113,7 @@ def training_step(self, batch: torch.Tensor, batch_idx: int) -> torch.Tensor:
             on_step=True,
             on_epoch=True,
             prog_bar=True,
-            logger=True,
+            # logger=True,
         )
 
         return loss