little fixes

neo_reader/multitask_dataset.py

@@ -28,7 +28,7 @@ class NpzRepresentation:
 
     def save_to_disk(self, data: tr.Tensor, path: Path):
         """stores this item to the disk which can then be loaded via `load_from_disk`"""
-        np.save(path, data.cpu().numpy(), allow_pickle=False)
+        np.save(path, data.cpu().detach().numpy(), allow_pickle=False)
 
     def plot_fn(self, x: tr.Tensor) -> np.ndarray:
         """very basic implementation of converting this representation to a viewable image. You should overwrite this"""
@@ -37,7 +37,7 @@ class NpzRepresentation:
         assert len(x.shape) == 3, x.shape # guaranteed to be (H, W, C) at this point
         if x.shape[-1] != 3: x = x[..., 0:1]
         if x.shape[-1] == 1: x = x.repeat(1, 1, 3)
-        x = x.nan_to_num(0).cpu().numpy() # guaranteed to be (H, W, 3) at this point hopefully
+        x = x.nan_to_num(0).cpu().detach().numpy() # guaranteed to be (H, W, 3) at this point hopefully
         _min, _max = x.min((0, 1), keepdims=True), x.max((0, 1), keepdims=True)
         if x.dtype != np.uint8: x = np.nan_to_num((x - _min) / (_max - _min) * 255, 0).astype(np.uint8)
         return x
@@ -119,7 +119,7 @@ class MultiTaskDataset(Dataset):
         for task_name in self.task_names:
             t = self.task_types[task_name]
             try:
-                t = t(task_name)
+                t = t(task_name) # hack for not isinstance(self.task_types, NpzRepresentation) but callable
             except Exception:
                 pass
             self._tasks.append(t)

neo_reader/neo_node.py

@@ -33,7 +33,7 @@ def _act_to_cmap(act_file: Path) -> np.ndarray:
     return rgb_colors
 
 class NEONode(NpzRepresentation):
-    """NEO nodes implementation in ngclib repository"""
+    """NEO nodes implementation"""
     def __init__(self, node_type: str, name: str):
         self.node_type = node_type
         self.name = name
@@ -45,11 +45,9 @@ class NEONode(NpzRepresentation):
     def load_from_disk(self, path: Path) -> tr.Tensor:
         data = np.load(path, allow_pickle=False)
         y = data if isinstance(data, np.ndarray) else data["arr_0"] # in case on npz, we need this as well
-        if y.shape[0] == 1: # pylint: disable=unsubscriptable-object
-            y = y[0] # pylint: disable=unsubscriptable-object
-        if len(y.shape) == 2:
-            y = np.expand_dims(y, axis=-1)
-        y[np.isnan(y)] = 0
+        y = y[0] if y.shape[0] == 1 else y # pylint: disable=unsubscriptable-object
+        y = np.expand_dims(y, axis=-1) if len(y.shape) == 2 else y
+        y[y == 0] = float("nan")
         return tr.from_numpy(y).float()
 
     @overrides
@@ -57,9 +55,9 @@ class NEONode(NpzRepresentation):
         return super().save_to_disk(data.clip(0, 1), path)
 
     def plot_fn(self, x: tr.Tensor) -> np.ndarray:
-        y = np.clip(x.cpu().numpy(), 0, 1)
+        y = np.clip(x.cpu().detach().numpy(), 0, 1)
         y = y * 255
-        y[y == 0] = 255
+        y[np.isnan(y)] = 255
         y = y.astype(np.uint).squeeze()
         y_rgb = self.cmap[y].astype(np.uint8)
         return y_rgb