Naming/Top1-accuracy inconsistency between LCE GitHub documentation and source files

[simonmaurer]

Hi

With respect to reporting new latency numbers for the QuickNet models on the Apple M1 platform I noticed the following:
In the source file of the QuickNet models the reported Top1-Accuracies are different than the one stated in the documentation.

Not to be picky, just to be consistent maybe we can update the naming to be consistent between documentation & source classes.

Is it correct that the following models coincide:
QuickNet (58.6%) <-> QuickNetSmall (59.4%)
QuickNetLarge (62.7%) <-> QuickNet (63.3%)
QuickNetXL (67%) <-> QuickNetLarge (66.9%)
?

[AdamHillier]

Hi @simonmaurer, thank you for pointing that out. You are absolutely right - we recently updated to new versions of the models, which are indeed named differently (and have slightly different accuracies). The version on the right is the 'correct' current version - we need to update our docs!

[simonmaurer]

perfect! I could update the ReadMe.md (as it is part of the GitHub repo) and add it in the next PR (macos_arm64 support) including numbers for the Apple mac mini (Apple M1, 1 and 4 threads)

[AdamHillier]

For now I've opened #663 - we need to collect new benchmark numbers before we can merge it. But it would be great to add in latencies for the M1 mac once you've added support for it!

[simonmaurer]

alright. added the benchmark results in the PR and as a comment in #663