search

BICLab / SpikeYOLO

Offical implementation of "Integer-Valued Training and Spike-Driven Inference Spiking Neural Network for High-performance and Energy-efficient Object Detection" (ECCV2024 Best Paper Candidate / Oral)
https://arxiv.org/abs/2407.20708
GNU Affero General Public License v3.0
97 stars 4 forks source link

A question about the training method using I-LIF #3

Closed caojiahui-ai closed 1 week ago

caojiahui-ai commented 1 month ago

Thanks for the new spike neurons, but I wonder why using I-LIF for training is called “direct training”. I-LIF still has MACs during training, which feels more like a compromise between ANN2SNN and directly trained SNN.

XinhaoLuo666 commented 1 week ago

Thanks for your question. Most of the methods for direct training of SNNs, such as LIF neurons, use the gradient substitution function method, which also has MAC in the training process, but does not include MAC in the inference process. It's the same with our I-LIF.