Error of Inverse Result is Large?

Question

koalazf99 opened this issue 5 years ago · comments

I use your command script to run a classification model and meet these 2 issues.

When the model hasn’t been trained, I test its inverse function. And the error of a (3x32x32 sized) picture is only about 0.001 when running 20 inverse iterations.

Then I try to load the model after 1 epoch, the reconstruction error is suddenly about 5.

I load the model after 50, 150, 200 epochs, but none of them can match the untrained model’s inverse error. After 200 epochs, for a (3x3x32 sized) picture, the smallest error is about 0.95.
When I use inverse iterations on the trained model, the reconstruction error rises when I use more inverse iterations. It’s strange because I think the more inverse iterations I use, the less inverse error I will get.

Is this result normal? This problem puzzles me a lot.

Jörn Jacobsen · Answer 1 · Thu Jun 13 2019 22:18:00 GMT+0800 (China Standard Time)

Can you redo this but with 20 instead of 5 power iterations:

Jörn Jacobsen · Answer 2 · Sun Jun 23 2019 00:15:00 GMT+0800 (China Standard Time)

@koalazf99 did increasing nr. of power iterations solve the problem for you?

Fan · Answer 3 · Sun Jun 23 2019 15:16:12 GMT+0800 (China Standard Time)

@koalazf99 did increasing nr. of power iterations solve the problem for you?

So sorry that I forgot to give a quick response.

Increasing the nr. of power iterations seems to be not useful in CIFAR dataset.

I later found that it may be the problem of loading the pretrained model and initializing the iResNet.

Anyway, the problem is solved now and thanks a lot for your help!