WIP: save work.

python/oddkiva/shakti/inference/yolo/darknet_layers.py

@@ -47,10 +47,16 @@ def __init__(self, in_channels, darknet_params: dict[str, Any], id: int):
             raise ValueError(f'No convolutional activation named {activation}')
         self.block.add_module(f'{activation}{id}', activation_fn);
 
-
     def forward(self, x):
         return self.block.forward(x)
 
+    def load_weights(self, weights_file: Path):
+        with open(weights_file, 'rb') as fp:
+            w_data = fp.read(conv.weight.shape.numel() * 4)
+            conv.weight.data.copy_(torch.from_numpy
+            conv.bias.data = fp.read(conv.bias.shape.numel() * 4)
+
+
 
 class MaxPool(nn.Module):
 

python/oddkiva/shakti/inference/yolo/darknet_model.py

@@ -13,6 +13,12 @@ def __init__(self, darknet_config: DarknetConfig):
 
         self.model = self.create_network(darknet_config)
 
+        self.major = None
+        self.minor = None
+        self.revision = None
+        self.seen = None
+        self.transpose = None
+
     def create_network(self, darknet_config: DarknetConfig):
         model = nn.ModuleList()
 
@@ -26,8 +32,9 @@ def create_network(self, darknet_config: DarknetConfig):
 
         return model
 
-    def load_weights(self, weights_file: Path):
-        pass
+    def load_convolutional_weights(self, conv, weights_file: Path):
+        with open(weights_file, 'rb') as fp:
+            fp.read(
 
     def save_weights(self, weights_file: Path)
         pass