LinglanZhao

icon company Shanghai Jiao Tong University

Results 5 comments of LinglanZhao

Testing problem

Same problem. Have you solved it yet?

Some questions about the code

https://github.com/icoz69/CaNet/blob/fdce9462d03ff93bc11f4795df94fbbf60b1c8ca/train.py#L103 Did you comment this line? When single GPU is available, using `model=nn.DataParallel(model,[0]`) may fix this problem.

Ask for dataset

You can download the data set manually in the following way: 1. Download CelebFaces Attributes Dataset (CelebA) from http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html using Google Drive or Baidu Drive. 2. Unzip CelebA/Img/img_align_celeba.zip and copy...

Feature value of x_ij in the egde matrix should be equal to the value of x_ji?

Same issue @li3cmz @khy0809. e_ijd seems different from e_jid in Equation (4) and (5) in the original paper, which conflicts with the strength of edge feature represented by the color...

pixel size of the miniImagenet in this work

Hi, I used conv4 protonet train_aug 5-way 5-shot on miniImagenet but the test accuracy was about 68%. How can you get the test result from 70.5 to 73.0? (How did...