Shu Kong

I see...your questions do not show up completely in email but in github. attribute features are concatenated with aesthetic features; but content features are multiplicatively gated with the content classification...

Hi, if you print out the layer's name, then you know the first layer's name which receives the input image. On Sun, Jul 14, 2019 at 9:04 AM RabiaSikandar wrote:...

Hi, thank you for your interest in our work, and apologize for not providing a readme. The installation follows the official caffe guide -- http://caffe.berkeleyvision.org/installation.html Just one thing to note...

I recall that label=-1 means open-set examples. Yes, the discriminator is a two-way classifier that discriminates between open- vs closed-set examples. On Sun, Jan 2, 2022 at 10:43 PM liMike1998...

hi, negative values means the attribute has a negatively effect on the image aesthetics. The annotation for each rater is in {-1,0,1}; given annotation of five raters, we average their...

In the following link, you can find read the ".mat" file for all the annotations, including raters' ID. Then you can generate training pairs in your way. https://drive.google.com/drive/folders/0BxeylfSgpk1MOVduWGxyVlJFUHM On Tue,...

Hi, thanks for your interest. I only upload a few examples for demonstration. If you want to test other images, you need to prepare the data in the same format...

As AADB is sort of small in scale, in our experiment, we first fine-tune the AlexNet backbone over AADB for aesthetic rating. Then we freeze AlexNet backbone and train only...

I believe this can be re-written into tensor operation. For example, for the following -- vgrid[:,0,:,:] = 2.0*vgrid[:,0,:,:]/max(W-1,1)-1.0 vgrid[:,1,:,:] = 2.0*vgrid[:,1,:,:]/max(H-1,1)-1.0 One way is to construction a tensor with H...

hi, thanks for your interest. If you use matlab, you can use the build-in function to calculate the correlation score. ref -- https://www.mathworks.com/help/stats/corr.html On Thu, Jul 6, 2017 at 2:30...