Teng Li

OK. I got it. You use other tricks like lr schedule?

Nice test @omoindrot . But I still have no idea about applying `DirectedInterleaveDataset` on raw msceleb1m dataset. I asked and commented on your answer on [stackoverflow](https://stackoverflow.com/questions/50550126/tf-data-api-how-to-produce-tensorflow-input-to-face-set-recognition). By the way, I...

@omoindrot since `choose_from_datasets` can randomly select element from dataset in datasets, we need shuffle dataset beforehand?

@omoindrot As my experimental result shows, it did not take the plate at the top but randomly take in each pile

@omoindrot Your triplet preparation code worked as num_labels=20000. However, when num_labels=40000, the error below occurred. It means this method cannot make triplets for a large number of classes? 

@MarcoForte Love to repeat your result by your training code.

@yutseng318 You may need to add the new line at the end of the original script.

@gaohongfein You built Tensorflow 1.9(currently tf-nightly-1.9) on cuda 9.2?

Have you tried building tf-nightly-1.9? I don't know which version of cuda can it build on? I built it on cuda-9.0 but Tensorflow did not use gpu acceleration.

I had the same problem