@tdeboissiere Line 93 in the densenet.py: > x = conv_factory(x, concat_axis, growth_rate, the input variable **growth_rate** shouldn't be changed to **nb_filter**? Since the nb_filter is growing in each layer in the dense block.