hejingwenhejingwen / adafm Goto Github PK

There are some problem on the subtraction of two images. how can I avoid it？
mse = np.mean((img1 - img2)**2)
ValueError: operands could not be broadcast together with shapes (330,502,3) (1348,2032,3)
It's basically in this position: "util.py", line 111

Filter visualization 是怎么实现的，论文中现实的是卷积核吗，大小为多少

I just ran 'python interpolate.py -opt options/test/test.json' without other operations.
My configuration is 2080ti.

Hi, you have done a great job!
Could you please share the supplemental material?
It is highly apprepriated if you can also share the PSNR distances of Modulation Testing (Sec. 4.3) on denoising and super resolution tasks. The performances look really good on the DeJEPG task.

Hi,i have one doubt about the paper and the corresponding code. in the paper ,fmid = f15 + lambda(g - I) * f15; but I think the code corresponds fmid = f15 + lambda*g * f15 ？I cannot fully understanding this .

According to your original "basic.json" file ,if i have a LR dataset with 120120 pixels images,and a HR dataset with 480480 pixels images; I need to modify the "crop_size" parameter from 0 to 4; but i find this is wrong ; i can't get any message from feeddata with same size 96*96 ; so how can i make it right?

Excuse me.During the test, the lamda parameter in the paper is interpolate_stride in the code? Why do I get the same result by adjusting from 0 to 1?

And have you tried using meta-learning to generate AdaFM instead of adjusting the weight of AdaFM?

Hi,

Thanks for your sharing.
I see there is not residual structure in the AdaFM module.
Why it returns self.transformer(x) + x instead of self.transformer(x) ?
Thanks.

The formulation in sec.3.3Comparison with BN layer,
where µ, σ are the mean and standard deviation of an input batch, γ, β are affine parameters.

Should it be sigma rather than gamma ?
Thanks~