Bill
Bill
为了使ann的激活范围和snn输出电流范围完全相同，采用了Optimized Potential Initialization for Low-latency Spiking Neural Networks. AAAI2022和Optimal ANN-SNN Conversion for High-accuracy and Ultra-low latency Spiking Neural Networks. ICLR2022的建模方法，将IF发放01脉冲调整为输出PSP，这样shortcut就不用处理了。更多内容敬请期待新教程
如果需要，请在问题中附上提供您的模型
I can't see your conversion code and model
我感觉create_conv_layers建出来的模型应该没问题啊
hi there, you can increase the timesteps to some bigger number like 500. MAXNORM and ROBUSTNORM are very primary methods yet useful. A small timestep may cause decrease in accuracy....
well, VGG-16, which pooling method do you choose?
great，change to avgpool and your problem may be solved! @sauravtii
try new algos I recommend you read [Optimized Potential Initialization for Low-latency Spiking Neural Networks](https://arxiv.org/abs/2202.01440)