NOTES.md

We report updated results of the provided pre-trained model on the auxilliary tasks (Table III):

(obj) cls, acc:	0.380
(obj) pos, err_mean:	2.495
(obj) pos, err_median:	1.745
(obj) pos, err_acc0.5:	0.145
(obj) pos, err_acc1.0:	0.304
(rel) dist, err_mean:	1.234
(rel) dist, err_median:	0.861
(rel) dist, err_acc0.5:	0.322
(rel) dist, err_acc1.0:	0.557

Note that these scores differ only slightly from the reported scores in the paper and, in particular, don't impact the conclusion of this experiment.