No accuracy is not included, since that is something that should be constant for each model across frameworks.

These benchmarks aim to highlight the performance differences in terms of speed/memory usage across frameworks and machine configurations.

There is also the practical hurdle that training imagenet models to maximum accuracy takes 1 week+.