VESPCN

# Pytorch's version
---
vespcn:
  depth: 3
  channel: 3
  scale: 4

batch: 16
patch_size: 64
lr: 1.0e-4
lr_decay:
  method: multistep
  decay_step: [200, 400]
  decay_rate: 0.1

# tensorflow's version
---
vespcn:
  depth: 3  # this `depth` is to inform graph builder
  beta: 1
  gamma: 0.01
  scale: 4
  channel: 3
  weight_decay: 0
batch: 4
patch_size: 96
depth: 3  # must be same as the model depth
lr: 1.0e-4
lr_decay:
  method: multistep
  decay_step: []
  decay_rate: 1

VDSR

vdsr:
  scale: 4
  channel: 3
  layers: 20
  upsample: true
  image_weight: 1
  feature_weight: 0  # perceptual loss

batch: 64
patch_size: 96
lr: 1.0e-2
lr_decay:
  method: multistep
  decay_step: [50, 75]
  decay_rate: 0.1

SRCNN

Trained on DIV2K dataset. Parameter file for this SRCNN

# srcnn 9-5-5
---
srcnn:
  layers: 3
  kernel:
    - 9
    - 5
    - 5
  scale: 4
  channel: 1

batch: 64
patch_size: 96
lr: 1.0e-2
lr_decay:
    method: multistep
    decay_step: [10000, 15000]
    decay_rate: 0.1

SPMC

# Detail-revealing Deep Video Super-resolution (ICCV 2017)
---
spmc:
  scale: 4
  channel: 3
  depth: 3
  # stage 1: optimize ME only
  # stage 2: optimize SR only
  # stage 3: jointly optimize
  stage: 3
  lambda1: 0.01
  lambda2: 0.01
  # add bicubic to output
  residual: false

batch: 16
patch_size: 96

SOFVSR

sofvsr:
  channel: 1
  scale: 4
  depth: 3

batch: 16
patch_size: 64
lr: 1.0e-4
lr_decay:
  method: multistep
  decay_step: [200, 400]
  decay_rate: 0.1

QPRN

qprn

update info

FRVSR

---
frvsr:
  scale: 4
  channel: 3
  depth: 7
  weights: [1, 1, 1.0e-3]

patch_size: 128
lr: 1.0e-4
lr_decay:
  method: multistep
  decay_rate: 0.5
  decay_step: [200, 300, 400]

ESRGAN

Trained on original author provided RRDB_PSNR.pth. 2000 * 200 steps.

esrgan:
  scale: 4
  in_nc: 3
  out_nc: 3
  nf: 64
  nb: 23
  gc: 32
  act_type: 'leakyrelu'
  mode: 'CNA'
  res_scale: 1
  upsample_mode: 'upconv'
  weights: [0.01, 1, 5.0e-3]
  patch_size: 128

batch_size: 16
lr: 1.0e-4
lr_decay:
  method: multistep
  decay_step: [250, 500, 1000, 1500]
  decay_rate: 0.5

ESPCN

Parameter for ESPCN

# espcn 5-3-3
---
espcn:
  layers: 3
  filters: [64, 32]
  kernel: [5, 3, 3]
  scale: 4
  channel: 1

batch: 64
patch_size: 96
lr: 1.0e-2
lr_decay:
    method: multistep
    decay_step: [10000, 15000]
    decay_rate: 0.1

DRRN

drrn:
  scale: 4
  channel: 3
  n_ru: 3
  n_rb: 3
  clip: 0.01