Arun Mallya

Yup, same error with VGG as the input CNN.

I trained from scratch using your code. Using the model saved after the first step, i tried to run the second step, which then failed giving the error above.

Yes, I got the error while training/fine-tuning. I used the following commands: 1) `th train_new.lua -max_iters 250000 -finetune_cnn_after 100000` 2) `th train_sc.lua -max_iters 150000 -start_from .t7 `

Technically, the loss would be different, but the gradients would be correct as `entropy(t)` doesn't contribute to gradient w.r.t `x`.

Hi, what command are you trying to run? The pretrained networks should be automatically loaded inside the modified networks (e.g. https://github.com/arunmallya/packnet/blob/master/src/networks.py#L32)

Avg pool does not have any weights, so I am not sure what error you're getting. The original VGG-16 definition does not have pooling as it only expects an input...

That's because the `arch_parameters` are not being copied onto every GPU. `DataParallel` only copies `parameters` and `buffers` of a module to all GPUs. In the above code, the `arch_parameters` are...