diff --git a/scripts/train_pytorch.py b/scripts/train_pytorch.py
index c7ddd2b595..5d1e27b534 100644
--- a/scripts/train_pytorch.py
+++ b/scripts/train_pytorch.py
@@ -374,9 +374,23 @@ def train_loop(config: _config.TrainConfig):
         batch_size = next(iter(sample_batch["image"].values())).shape[0]
         for i in range(min(5, batch_size)):
             # Concatenate all camera views horizontally for this batch item
-            # Convert from NCHW to NHWC format for wandb
-            img_concatenated = torch.cat([img[i].permute(1, 2, 0) for img in sample_batch["image"].values()], axis=1)
+            frames = []
+            for img in sample_batch["image"].values():
+                frame = img[i]
+                # Detect format: NCHW has channels (3) as first dim, NHWC has channels as last dim
+                if frame.shape[0] == 3:
+                    # NCHW format -> convert to NHWC
+                    frame = frame.permute(1, 2, 0)
+                # else: already NHWC format, no conversion needed
+                frames.append(frame)
+            # Concatenate along width dimension (dim=1 for NHWC)
+            img_concatenated = torch.cat(frames, dim=1)
+            # Convert to numpy and ensure values are in [0, 1] range for wandb
             img_concatenated = img_concatenated.cpu().numpy()
+            if img_concatenated.min() < 0:
+                # Values in [-1, 1] range, rescale to [0, 1]
+                img_concatenated = (img_concatenated + 1) / 2
+            img_concatenated = np.clip(img_concatenated, 0, 1)
             images_to_log.append(wandb.Image(img_concatenated))
 
         wandb.log({"camera_views": images_to_log}, step=0)