search

jinningli / IP-FSRGAN

Source code for the paper "ID Preserving Face Super-ResolutionGenerative Adversarial Networks"
4 stars 2 forks source link

Please Guide How to train this model with reference to dataset #2

Open Basir-mahmood opened 3 years ago

Basir-mahmood commented 3 years ago

Please guide which dataset is used and what are those lists that is referred for the training of the model. Moreover, can you please describe how the triplet images are developed for the training ?

jinningli commented 3 years ago

Hi,

I am sorry that the dataset used in the paper is not published, because the data was stored on the server of my cooperating company. The path lists are referred to those servers, which I cannot get access now.

For the super-resolution module, the low-resolution data is generated by down-sampling with opencv. For the ID preserving module, the positive and negative images are sampled from the dataset, you can download the dataset I mentioned in the paper and you will find, for example, a series of face images of person A. The face images of the same person can be sampled as positive pairs.

Thanks