search

putshua / ANN_SNN_QCFS

Code for paper "Optimal ANN-SNN conversion for high-accuracy and ultra-low-latency spiking neural networks"
18 stars 5 forks source link

Converting BN layers #6

Open MusheerAbdullah opened 6 months ago

MusheerAbdullah commented 6 months ago

Dear Sir, Hope you are doing well I wonder, how do you deal with converting BN layer in your code, most of SOTAs absorb the BN parameters to weight and bias of Conv layers, but I cannot see this on your code, How do you convert the BN layers?

Thanks in advance

putshua commented 6 months ago

The BN can be absorbed into conv layers. You can refer to the old version https://github.com/putshua/SNN_conversion_QCFS, utils.py to find the code. Based on my experiments, there is very litte change in performance before and after the modification.