WIP DVS - DVS node not given

dvs_input (bool) arg passed down to collect_dynapcnn_layer_info() to raise error if edge type involving DVS is not found when dvs_input == True.

sinabs/backend/dynapcnn/dynapcnn_network.py

@@ -69,15 +69,15 @@ def __init__(
             batch_size = sinabs.utils.get_smallest_compatible_time_dimension(snn)
         # computational graph from original PyTorch module.
         self._graph_extractor = GraphExtractor(
-            snn, torch.randn((batch_size, *self.input_shape))
+            snn, torch.randn((batch_size, *self.input_shape), self.dvs_input)
         )  # needs the batch dimension.
 
         # Remove nodes of ignored classes (including merge nodes)
         self._graph_extractor.remove_nodes_by_class(DEFAULT_IGNORED_LAYER_TYPES)
 
         # Module to execute forward pass through network
         self._dynapcnn_module = self._graph_extractor.get_dynapcnn_network_module(
-            discretize=discretize, weight_rescaling_fn=weight_rescaling_fn
+            discretize=discretize, weight_rescaling_fn=weight_rescaling_fn, dvs_input=self.dvs_input
         )
         self._dynapcnn_module.setup_dynapcnnlayer_graph(index_layers_topologically=True)
 

sinabs/backend/dynapcnn/nir_graph_extractor.py

@@ -105,7 +105,7 @@ def indx_2_module_map(self) -> Dict[int, nn.Module]:
         return {n: module for n, module in self._indx_2_module_map.items()}
 
     def get_dynapcnn_network_module(
-        self, discretize: bool = False, weight_rescaling_fn: Optional[Callable] = None
+        self, discretize: bool = False, weight_rescaling_fn: Optional[Callable] = None, dvs_input: bool = False
     ) -> DynapcnnNetworkModule:
         """ Create DynapcnnNetworkModule based on stored graph representation
 
@@ -117,6 +117,7 @@ def get_dynapcnn_network_module(
             weights to dynapcnn. Set to `False` only for testing purposes.
         - weight_rescaling_fn (callable): a method that handles how the re-scaling factor for one or more `SumPool2d` projecting to
             the same convolutional layer are combined/re-scaled before applying them.
+        - dvs_input (bool): wether or not dynapcnn receive input from its DVS camera.
 
         Returns
         -------
@@ -129,6 +130,7 @@ def get_dynapcnn_network_module(
             edges = self.edges,
             nodes_io_shapes=self.nodes_io_shapes,
             entry_nodes=self.entry_nodes,
+            dvs_input=dvs_input,
         )
 
         # build `DynapcnnLayer` instances from mapper.
@@ -236,8 +238,8 @@ def _add_dvs_node(self, dvs_input_shape: Tuple[int, int]) -> None:
         ----------
         - dvs_input_shape (tuple): Shape of input in format `(height, width)`. 
         """
-        # @TODO - not considering pooling after the DVSLayer yet.
-        # @TODO - does self._entry_nodes need to have the index of the DVS node?
+
+        # [] @TODO - not considering pooling after the DVSLayer yet.
 
         # add name entry for node 'dvs'.
         self._name_2_indx_map['dvs'] = len(self._name_2_indx_map)
@@ -246,6 +248,8 @@ def _add_dvs_node(self, dvs_input_shape: Tuple[int, int]) -> None:
         # set DVS node as input to each entry node of the graph.
         self._edges.update({(self._name_2_indx_map['dvs'], entry_node) for entry_node in self._entry_nodes})
 
+        # [] @TODO - all indexes in 'self._entry_nodes' are no longer entry nodes of the network since a DVS layer is being added.
+
     def _need_dvs_node(self, model: nn.Module, dvs_input: bool) -> bool:
         """ Returns whether or not a node will need to be added to represent a `DVSLayer` instance. A new node will have 
         to be added if `model` does not start with a `DVSLayer` instance and `dvs_input == True`.

sinabs/backend/dynapcnn/sinabs_edges_handler.py

@@ -25,6 +25,7 @@ def collect_dynapcnn_layer_info(
     edges: Set[Edge],
     nodes_io_shapes: Dict[int, Dict[str, Tuple[Size, Size]]],
     entry_nodes: Set[int],
+    dvs_input: bool,
 ) -> Dict[int, Dict]:
     """Collect information to construct DynapcnnLayer instances.
 
@@ -37,10 +38,11 @@ def collect_dynapcnn_layer_info(
 
     Parameters
     ----------
-    indx_2_module_map (dict): Maps node IDs of the graph as `key` to their associated module as `value`
-    edges (set of tuples): Represent connections between two nodes in computational graph
-    nodes_io_shapes (dict): Map from node ID to dict containing node's in- and output shapes
-    entry_nodes (set of int): IDs of nodes that receive external input
+    - indx_2_module_map (dict): Maps node IDs of the graph as `key` to their associated module as `value`
+    - edges (set of tuples): Represent connections between two nodes in computational graph
+    - nodes_io_shapes (dict): Map from node ID to dict containing node's in- and output shapes
+    - entry_nodes (set of int): IDs of nodes that receive external input
+    - dvs_input (bool): wether or not dynapcnn receive input from its DVS camera.
 
     Returns
     -------
@@ -62,6 +64,11 @@ def collect_dynapcnn_layer_info(
             "None such weight-neuron pair has been found in the provided network."
         )
 
+    if not any(edge in edges_by_type for edge in ["dvs-weight", "dvs-pooling"]) and dvs_input:
+        raise InvalidGraphStructure(
+            "DVS camera is set selected for usage (dvs_input == True) but edge type involving it has not been found."
+        )
+
     # Dict to collect information for each future dynapcnn layer
     dynapcnn_layer_info = dict()
     # Map node IDs to dynapcnn layer ID