数据一样，绘制的图片不同

[MaWenyun]

trafficstars

您好，请问您论文中的在不同计算比 r 情况下的端到端延迟的那张图。是如何绘制的呢？尤其是横坐标的设置。以及根据您论文中的公式。得出 ：End to End lantency = (r +1) 服务器端延迟+传输工作量/R 是否正确？

[hou-yz]

thank you for your interest.

this project was a while ago so I'm not 100% sure about the details.

if I recall correctly, in Fig. 3 and Fig. 4, the x-axis should denote the layer ID from VGG-16. Fig. 3 shows the details about the computation and communication workload needed for hidden vectors at each layer, and Fig. 4 shows the latency & accuracy for different partition points between mobile phone and cloud server device. both figures show the original VGG as baseline, the 1-step pruning approach, and the 2-step pruning approach.

the total (end-to-end) latency should be $t^mobile+t^transmission+t^device$, which can be found in Algorithm 1 line 9.

hope this helps.