What does the confident map mean?

[John1231983]

It is not a bug. I just want to ask this question to make more clearly understand.
Your discriminator network produces a spatial map by using fully convolution layer.

1. Does the spatial map same as the confident map? Or confident map generates by unlabeled data?
2. Does the value of the spatial map (from D network) in range 0 to 1 or just {0,1}?

[hfslyc]

The output of D is a probability map with values ranges from 0-1. When the input is unlabeled data, we use the D output map as the indicator for semi-supervised learning.

[John1231983]

Thanks for your reply. So, what does confident map mean( in question 1)? Is it just ouput of D?

[chuanruihu]

Hi, when I train the model, I find that the loss_seg is always about 1.5 shock. The loss_adv_p is up to a few dozen(80). The loss_D is about 0.1. I just train the model without pretrained model. Your code train.py line 194 to line 207 I just log off to train the new model. Can you tell me how to train a new model. Thank you !

[John1231983]

Me too. I also interested how to train the network from scratch.

[rhp228228]

Me too. The code may be has some wrong. I am not clearly. I also want to know how to train the network from scratch.

[hfslyc]

Hi all, I'm investigating the problem. Something might go wrong when I was cleaning the code.

[chuanruihu]

Thank you! Your work is really interesting . I look forward you can check the code quickly, and can write the README detailedly. thank you.

[hfslyc]

@chuanruihu how do you train the model without pretrained model? The training process should always start from a imagenet pretrained model. Otherwise, it will not converge.

[chuanruihu]

@hfslyc Hi, I train the model without pretrained model, is just log off(注释) line194 to line207 in train.py. After you reply ,I konw your code is must start from a imagenet pretained model. Thank you for your reply.

Second! when I just trained the model, I find that loss_semi is always in {0.001, 0.002, 0.003...}. I don't know
why the loss_semi is always this number. and the loss_semi is what ?

[hfslyc]

@chuanruihu please refer to our paper for detailed description for semi-supervised loss

[hfslyc]

Hi all, if there is no further issue. I'm closing this one. Again, thank you for interested in our work.

[John1231983]

@hfslyc : Sorry for reply late. Could you guide me how could we obtain the figure of confidence map in your figure? Just feed the prediction to the trained weighted of D network? Do we need to use argmax()