unreasonable accuracy numbers for vgg8

[sbastani]

I have downloaded the weight parameters of vgg8 from the link provided in the manual.
When I run the example vgg8 use-case in the manual using MNSIM 2.0 with the default hardware settings (e.g. no device non-ideality) it shows original accuracy of 10.8% (more specifically, Original accuracy: 0.10818181818181818) and PIM-based computing accuracy: 0.10818181818181818. I ran the simulation on CPU.
My questions are:
1- What does the original accuracy mean? Is it pure software based accuracy?
2- Is the low accuracy of the vgg8, as described above, expected?

[sbastani]

Will this repo be maintained, and issues will be addressed or is it just a one-time publication-purposed repo?

[frankinwi]

@sbastani Same problem. @Zhu-Zhenhua Can the author provide more details about how to reproduce the results listed in the paper, e.g., 90.18% accuracy for the VGG-8 model with R=8?