Skip to content

Latest commit

 

History

History
55 lines (39 loc) · 1.05 KB

File metadata and controls

55 lines (39 loc) · 1.05 KB

Notes on Different Models and Datasets

Datasets

  • PIRM
    • 100 Images
    • Resolution ~400x600
Name n Resolution
PIRM 100 ~400x600
T91 91 ~150x200

Experiments with ResNet

Models

Name # Resnet Blocks Trainable Parameters
ResSR1 1 23.523
ResSR2 2 97.315
ResSR3 3 392.355

Training

Dataset: PIRM, patches 224x224, 1046 patches, 4x upscaling, batch size 64

100 Epochs, no LR Scheduler

Name Set5 Set14
Model1 22.791 21.517
Model2 21.460 20.703
Model3 19.529 18.965

Experiments resnet with maxpooling vs vgg

Dataset: T91, patches 31x31, 4x upscaling

100 Epochs, no LR Scheduler

Name Set5 Set14
ResSR1_n 23.391 21.933
VGG 25.034 23.024
VGGSR2 22.511 21.059
VGGSR0 25.519 23.496

final experiments

Name Set5 Set14
VGGSR 24.720 22.866
VGGSR2
VGGSR0