Which dataset is it in the demo?

[hbk16]

Which dataset is it in the demo? And where to download it? THX.

[zengyu714]

Here it is http://ijoint.istb.unibe.ch/challenge/index.html <http://ijoint.istb.unibe.ch/challenge/index.html>. But this challenge seems not open currently. You can just find another dataset to try some interesting algorithms.

…

On Jul 8, 2018, at 20:27, hbk16 ***@***.***> wrote: Which dataset is it in the demo? And where to download it? THX. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ATbNA-O3znfELCGPEzXvBiB36KD5gEzkks5uEfqVgaJpZM4VGtOq>.

[hbk16]

Thanks.
I've noticed that labels in training are converted to type LongTensor, but wouldn't it save lots of memories if using ShortTensor or CharTensor instead in dice loss?

[zengyu714]

You’re right in some way. Just when I wrote it, the loss function like `nll_loss <https://pytorch.org/docs/master/nn.html#nll-loss>` required specific LongTensor : ) I don’t know how it’s going now. This example<https://github.com/pytorch/pytorch/blob/e8536c08a16b533fe0a9d645dd4255513f9f4fdd/torch/nn/functional.py#L1412> in the documentation says it might not be necessary at present. (Since the `target` would be `torch.float32`) Anyway, just try more and choose the better one.

…

On Jul 8, 2018, at 23:17, hbk16 ***@***.***> wrote: Thanks. I've noticed that labels in training are converted to type LongTensor, but wouldn't it save lots of memories if using ShortTensor or CharTensor instead in dice loss? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ATbNA1HvvwPfVgJfx4jG-xpqF0OobrJxks5uEiKfgaJpZM4VGtOq>.