RePr

RePr copied to clipboard

RuntimeError: sizes must be non-negative

1

Hello, when I run the RePr training scheme, the result shows that 'RuntimeError: sizes must be non-negative', so I am wondering can you help me with that? Thank you very...

niuzchinanick

Ａ question about reinitialize

9

Hi, I feel a little confused about reinitialize. After reinitialize, the weights of the network have changed immediately. The accuracy may go down at the reinitializing point. Could you tell...

Yanxingang

Question about reinitialize

6

Hi, I feel a little confused about reinitialize. You have set `W.data[mask[name]]` to zero in pruning, however in reinitialize, you don't recover the corresponding weight to the dropped filter. In...

philokey

Missing abs() when calculating the inter-filter orthogonality?

12

https://github.com/siahuat0727/RePr/blob/4fa1b8d30da1d9f88ccc29fe3ab1711811ef093d/main.py#L148 In paper it seems need an abs() op here.

qinjian623