In the original activations/core.py: ActivationsHelper._get_activations_batched, the np.concatenate for accumulating output features will drastically slow down the program (as the accumulated result becomes larger). Now I fix it by simply using list to keep track of all outputs and concatenate them at the very end.
In the original activations/core.py: ActivationsHelper._get_activations_batched, the np.concatenate for accumulating output features will drastically slow down the program (as the accumulated result becomes larger). Now I fix it by simply using list to keep track of all outputs and concatenate them at the very end.