search

fangpin / siamese-pytorch

Implementation of Siamese Networks for image one-shot learning by PyTorch, train and test model on dataset Omniglot
279 stars 59 forks source link

Testing returns all cases matching with batch size one #3

Closed Sumes closed 5 years ago

Sumes commented 5 years ago

In all the case we are taking L1 distance of siamese fc outputs which leads to negative values. When taking sigmoid on this it always gives 0. eventually leading all values predicted correctly

fangpin commented 5 years ago

@Sumes

In all the case we are taking L1 distance of siamese fc outputs which leads to negative values. When taking sigmoid on this it always gives 0. eventually leading all values predicted correctly

What do you mean? There is no negative value since I use absolute L1 distance. See model.py https://github.com/fangpin/siamese-pytorch/blob/cef7cec2e7928ad4002e4813597a98c02fd7201d/model.py#L35

Sumes commented 5 years ago

By getting all predictions as 0 means that, even dissimilar images are also getting matched as same

On Sat, Oct 27, 2018 at 6:23 PM Pin,Fang notifications@github.com wrote:

Closed #3 https://github.com/fangpin/siamese-pytorch/issues/3.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/fangpin/siamese-pytorch/issues/3#event-1930300680, or mute the thread https://github.com/notifications/unsubscribe-auth/AoXVS7iy29bO4f8ASPrPbXFMENHuBBpgks5upFdGgaJpZM4WuboJ .