Hi,I have a problem?
BCWang93 opened this issue · comments
When I test the trained model use the test data or the train data ,the output is only zero.Anyone can help me to solve this?Thanks!
the output like this:
[[0 0 0 ..., 0 0 0]
[0 0 0 ..., 0 0 0]
[0 0 0 ..., 0 0 0]
...,
[0 0 0 ..., 0 0 0]
[0 0 0 ..., 0 0 0]
[0 0 0 ..., 0 0 0]]
Is this the result with VOC dataset?
On Thu, May 2, 2019 at 12:18 Bcw93 @.***> wrote: When I test the trained model use the test data or the train data ,the output is only zero.Anyone can help me to solve this?Thanks! the output like this: [[0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] ..., [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0]] — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#117>, or mute the thread https://github.com/notifications/unsubscribe-auth/ABA4LE2MJS2ZVDHQCOW5TXTPTLEY5ANCNFSM4HJ5NTXQ .
-- Kentaro Wada / 和田 健太郎 https://wkentaro.com http://wkentaro.com
yes,I use the voc2012 dataset,and I don't know why I has this problem?
Is this the result with VOC dataset?
On Thu, May 2, 2019 at 12:18 Bcw93 @.***> wrote: When I test the trained model use the test data or the train data ,the output is only zero.Anyone can help me to solve this?Thanks! the output like this: [[0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] ..., [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0]] — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#117>, or mute the thread https://github.com/notifications/unsubscribe-auth/ABA4LE2MJS2ZVDHQCOW5TXTPTLEY5ANCNFSM4HJ5NTXQ .
-- Kentaro Wada / 和田 健太郎 https://wkentaro.com http://wkentaro.com
Btw,I trained the FCN32 network.
Hmm, maybe because of the version of pytorch. The code in this repo is written with pytorch 0.2.0: https://github.com/wkentaro/pytorch-fcn#requirements
Maybe 1.0.0 doesn't work without changes.
Hmm, maybe because of the version of pytorch. The code in this repo is written with pytorch 0.2.0: https://github.com/wkentaro/pytorch-fcn#requirements
Maybe 1.0.0 doesn't work without changes.
Hi,I have another question!why the loss is so big?
It is big if you don't average with the number of pixels.
pytorch-fcn/torchfcn/trainer.py
Lines 112 to 113 in 97189cb