Results about VehicleID Dataset about deep-metric-learning-baselines

[Confusezius]

Hi there!

Since Vehicle-ID is currently not as widely used and not easily accessible (and quite large), I haven't run replicable fixed seed runs (yet) on this dataset, hence why they haven't been included.

The training code/dataloader is set up to work directly with this dataset though, and in sample_training_runs.sh you have (commented out) sample training commands to achieve good results on this dataset.

As a reference however, for the small and large test set, ResNet with margin loss currently achieves about R@1/R@5 of ~85/~92 and ~80/~89 respectively, with NPair and semihard trailing a bit behind.

Hope this helps! If I have time in the future, I will include these results as well.