3 ms·
It is not necessary to train things from scratch; you take the largest imagenet model available and fine tune it for the task. This way it reuses much of the lo
by cmarschner 9y ago
It is not necessary to train things from scratch; you take the largest imagenet model available and fine tune it for the task. This way it reuses much of the lower layers the have seen lots of data.