bugfix: Generator for AMD

lib/training/generator.py

@@ -71,7 +71,7 @@ class DataGenerator():
         self._batch_size = batch_size
 
         self._process_size = max(img[1] for img in model.input_shapes + model.output_shapes)
-        self._output_sizes = [shape[1] for shape in model.output_shapes if shape[-1] != 1]
+        self._output_sizes = self._get_output_sizes(model)
 
         self._coverage_ratio = model.coverage_ratio
         self._color_order = model.color_order.lower()
@@ -103,6 +103,27 @@ class DataGenerator():
             channels += len(mults)
         return channels
 
+    def _get_output_sizes(self, model: "ModelBase") -> List[int]:
+        """ Obtain the size of each output tensor for the model.
+
+        Parameters
+        ----------
+        model: :class:`~plugins.train.model.ModelBase`
+            The model that this data generator is feeding
+
+        Returns
+        -------
+        list
+            A list of integers for the model output size for the current side
+        """
+        out_shapes = model.output_shapes
+        split = len(out_shapes) // 2
+        side_out = out_shapes[:split] if self._side == "a" else out_shapes[split:]
+        retval = [shape[1] for shape in side_out if shape[-1] != 1]
+        logger.debug("side: %s, model output shapes: %s, output sizes: %s",
+                     self._side, model.output_shapes, retval)
+        return retval
+
     def minibatch_ab(self, do_shuffle: bool = True) -> Generator[BatchType, None, None]:
         """ A Background iterator to return augmented images, samples and targets.
 
@@ -313,7 +334,6 @@ class DataGenerator():
         batch = self._buffer()
         self._crop_to_coverage(filenames, raw_faces, detected_faces, batch)
         self._apply_mask(detected_faces, batch)
-
         return self.process_batch(filenames, raw_faces, detected_faces, batch)
 
     def process_batch(self,