Add gamma and beta attrs for batchnormalization. Account for no scale and offset

Author: jicampos

hls4ml/converters/keras/core.py

@@ -129,6 +129,9 @@ def parse_batchnorm_layer(keras_layer, input_names, input_shapes, data_reader):
     elif len(input_shapes[0]) == 4:
         layer['n_filt'] = input_shapes[0][3]
 
+    layer['use_gamma'] = False if keras_layer['config']['scale'] == False else True
+    layer['use_beta'] = False if keras_layer['config']['center'] == False else True
+
     return layer, [shape for shape in input_shapes[0]]
 
 

hls4ml/converters/pytorch/core.py

@@ -61,6 +61,11 @@ def parse_batchnorm_layer(pytorch_layer, layer_name, input_shapes, data_reader,
     #batchnorm para
     layer['epsilon'] = pytorch_layer.eps
 
+    if pytorch_layer.affine is False:
+        layer['use_gamma'], layer['use_beta'] = False, False
+    else:
+        layer['use_gamma'], layer['use_beta'] = True, True
+
     in_size = 1
     for dim in input_shapes[0][1:]:
         in_size *= dim

hls4ml/model/layers.py

@@ -800,8 +800,8 @@ def initialize(self):
         dims = inp.dim_names
         self.add_output_variable(shape, dims)
 
-        gamma = self.model.get_weights_data(self.name, 'gamma')
-        beta = self.model.get_weights_data(self.name, 'beta')
+        gamma = 1 if self.attributes['use_gamma'] is False else self.model.get_weights_data(self.name, 'gamma')
+        beta = 0 if self.attributes['use_beta'] is False else self.model.get_weights_data(self.name, 'beta')
         mean = self.model.get_weights_data(self.name, 'moving_mean')
         var = self.model.get_weights_data(self.name, 'moving_variance')