UMSN Files
This commit is contained in:
@@ -17,7 +17,7 @@ We propose a novel multi-stream architecture and training methodology that explo
|
||||
1. Download test datasets provided the authors of Ziyi et al.
|
||||
- https://sites.google.com/site/ziyishenmi/cvpr18_face_deblur
|
||||
2. run test_data_generation.m
|
||||
- It renames the files counting from example 000001.png
|
||||
- It renames the files counting from 1, for example 000001.png
|
||||
3. python test_face_deblur.py --dataroot ./facades/github/ --valDataroot <path_to_test_data> --netG ./pretrained_models/Deblur_epoch_Best.pth
|
||||
|
||||
## To train UMRL:
|
||||
|
||||
Reference in New Issue
Block a user