noelcodella / tripletloss-keras-tensorflow Goto Github PK

Thank you for providing such a valuable code for teaching tripletloss.
I read the code, generated .txt, and ran it successfully.
However, I got a strange problem.
When I feed "the training data" to do "mode.fit", the returned loss value say loss_1 = 0.4684.
But, I feed exactly the same data into "model.evaluate" (right after the above code), I got loss_2 = 1.0.
The loss_1 should be equal (or similar) to loss_2, but It seems not like that.
I had tried for a couple of days to figure out what happened, but I still not found the answer.
It will be appreciated if you could provide some comments, Thanks a lot.

Hey, thanks for triple loss example!
Could you explain why LR_SGD is used? Could I use any Keras-provided optimizers instead?
Greetings

I use triple loss between data of two modalities to reduce the distance between different modalities of the same class and increase the distance between different modalities of different class. But when I use batch_all loss, the valid set loss has not changed; now using hard_loss, the valid set loss still has not changed. What is the reason? I found some answers that triplet is difficult to converge. What do you do to deal with triplet loss convergence?

Hi! Two questions:

• Is this code adapt to multiclass instances or only for binary cases?
• Is possible to have any txt example?
  thanks!

Hello! First of all, thanks for releasing this didactic code.

Can you make available some example *.txt anchor, positive and negative pairs to run a small test?