Refactor examples and tests for improved clarity and coverage

- Removed unnecessary print statements from plot_channel_aware_base_model.py to streamline output.
- Updated plot_fec_decoders_tutorial.py and plot_fec_encoders_tutorial.py by removing redundant section headers for better readability.
- Added comprehensive tests for Structural Similarity Index Measure (SSIM) and Multi-Scale SSIM (MS-SSIM) metrics, ensuring various scenarios are covered.
- Introduced Error Vector Magnitude (EVM) metric tests, including basic computation, edge cases, and stateful methods.
- Enhanced Bit Error Rate (BER) tests to cover stateful methods and complex data handling.
- Improved test coverage for Signal-to-Noise Ratio (SNR) and Block Error Rate (BLER) metrics, addressing edge cases and ensuring robustness.

Author: selimfirat

kaira/metrics/image/ssim.py

@@ -61,6 +61,13 @@ def forward(self, x: Tensor, y: Tensor, *args: Any, **kwargs: Any) -> Tensor:
         """
         # Note: *args and **kwargs are not directly used by self.ssim call here
         # but are included for interface consistency.
+
+        # Handle empty tensors gracefully
+        if x.numel() == 0 or y.numel() == 0:
+            # Return empty tensor with appropriate shape
+            batch_size = x.shape[0] if x.numel() >= 0 else 0
+            return torch.tensor([], device=x.device, dtype=x.dtype).view(batch_size)
+
         values = self.ssim(x, y)
 
         # Apply reduction if specified
@@ -155,6 +162,12 @@ def forward(self, x: torch.Tensor, y: torch.Tensor, *args: Any, **kwargs: Any) -
         # Note: *args and **kwargs are not directly used here
         # but are included for interface consistency.
 
+        # Handle empty tensors gracefully
+        if x.numel() == 0 or y.numel() == 0:
+            # Return empty tensor with appropriate shape
+            batch_size = x.shape[0] if x.numel() >= 0 else 0
+            return torch.tensor([], device=x.device, dtype=x.dtype).view(batch_size)
+
         # Use torchmetrics MS-SSIM implementation
         values = self.ms_ssim(x, y)
 
@@ -178,6 +191,10 @@ def update(self, preds: torch.Tensor, targets: torch.Tensor, *args: Any, **kwarg
             *args: Variable length argument list passed to forward.
             **kwargs: Arbitrary keyword arguments passed to forward.
         """
+        # Handle empty tensors gracefully
+        if preds.numel() == 0 or targets.numel() == 0:
+            return  # Skip update for empty tensors
+
         values = self.forward(preds, targets, *args, **kwargs)  # Pass args/kwargs
         if values.numel() == 0:
             return  # Avoid updating with empty values