How to test unpaired dataset?

[JustinAsdz]

Hello, recently I'm reproducing your work, and have successfully completed the training process. When Test, I have got a problem.

i) The json file of test says the model = "sr", which means it lacks the ability to test datasets like Set5, BSD100 according to the SR_model.py , How to deal with such need? I wander If there are built-in method about this situation or just resize the imgs by myself? ii) In the json file , has a key named: pretrain_model_D_wlt, where is the weight_map comes from?

Thanks for your work and looking forward to your quick reply : )

[DizzyYunxuan]

The SRN code is built from mmsr repo. You can find more details on their Github pages.

i):

• Model. model='sr' means using SR_model.py in path SRN/models/ as model frameworks.
• Built-in downsample method. If you want to super-resolve the bicubic downsample LR, you only need to specify the HR path, leave LR path being 'null', and specify the test mode: LRHR. The data loader will downsample the HR by bicubic interpolation during testing.

ii):

• KeyName. The key pretrain_model_D_wlt denotes the pre-trained wavelet discriminator to be loaded.
• Domain-distance Map. Specify the key dataroot_fake_weights to the weight map path. The data loader LRHR_wavelet_unpair_fake_weights_EQ will load the weights map during training.