CSPNet: A New Backbone that can Enhance Learning Capability of CNN(2019)
Abstract1Introduction2Relatedworkformerwork3Method3.1CrossStagePartialNetwork3.2ExactFusionModel4Experiments5Conclusion原文链接源代码-梯度信息重用(有别于冗余的梯度信息)可以减少计算量和内存占用提高效率,但会降低精度(可能是因为重用梯度信息带来的近似误差