(Fork of BasicSR joeyballentine/traiNNer-redux) Open Source Image and Video Restoration Toolbox for Super-resolution, Denoise, Deblurring, etc. Currently, it includes EDSR, RCAN, SRResNet, SRGAN, ESRGAN, EDVR, BasicVSR, SwinIR, ECBSR, OmniSR, HAT, GRL, A-ESRGAN, etc. Also support StyleGAN2, DFDNet.

NEW ADD ARCH SUPPORT

• [ESRGAN 8X]
• Omnisr
  • The arch implementation of Omnisr is from Omnisr. The LICENSE of Omnisr is [Apache License 2.0]. The LICENSE is included as LICENSE_Omnisr.
• HAT
  • The arch implementation of HAT is from HAT. The LICENSE of HAT is [MIT License]. The LICENSE is included as LICENSE_HAT.
• GRL
  • The arch implementation of GRL is from GRL. The LICENSE of GRL is [MIT License].
• ESWT
  • The arch implementation of ESWT is from ESWT. The LICENSE of ESWT is [MIT License].
• SRFormer
  • The arch implementation of SRFormer is from SRFormer. The LICENSE of SRFormer is [Apache License 2.0].
• A-ESRGAN
  • The arch implementation of A-ESRGAN is from A-ESRGAN. The LICENSE of A-ESRGAN is [BSD 3-Clause "New" or "Revised" License].

NEW FEATURE SUPPORT

• ContextualLoss weight
• amp support

  try_autoamp_g: True # enable amp Automatic mixed precision for network_g. if loss inf or nan or error just set to False
  try_autoamp_d: True # enable amp Automatic mixed precision for network_d. if loss inf or nan or error just set to False
  

TIPS

• PairedImageDataset set high_order_degradation : False
• RealESRGANDataset set high_order_degradation : True
• To use Automatic mixed precision, edit the yml files in options\train\ESWT\
• e.g. if you want use omnisr-gan with amp, just edit the 4x_trainESWTGAN_SR_scratch-DIV2K.yml 's following part

name: train_ESWTGANModel_SRx4_scratch_P48W8_DIV2K_500k_B4G8
try_autoamp_g: True # enable amp Automatic mixed precision for network_g. if loss inf or nan or error just set to False
try_autoamp_d: True # enable amp Automatic mixed precision for network_d. if loss inf or nan or error just set to False
network_g:
  type: OmniSRNet
  num_in_ch: 3
  num_out_ch: 3
  num_feat: 64
  upsampling: 4
  res_num: 5
  block_num: 1
  bias: True
  block_script_name: OSA
  block_class_name: OSA_Block
  window_size: 8
  pe: True
  ffn_bias: True

• If you want to use aesrgan's network_d for other network_d. you should edit the 4x_train_multiaesrgan_plus.yml
• e.g. if you want use omnisr as network_g and multiscale as network_d, just edit the 4x_train_multiaesrgan_plus.yml 's following part

# network structures
# network_g:
#   type: RRDBNet
#   num_in_ch: 3
#   num_out_ch: 3
#   num_feat: 64
#   num_block: 23
#   num_grow_ch: 32

network_g:
  type: OmniSRNet
  num_in_ch: 3
  num_out_ch: 3
  num_feat: 64
  upsampling: 4
  res_num: 5
  block_num: 1
  bias: True
  block_script_name: OSA
  block_class_name: OSA_Block
  window_size: 8
  pe: True
  ffn_bias: True

#network_d:
#  type: UNetDiscriminatorAesrgan
#  num_in_ch: 3
#  num_feat: 64
#  skip_connection: True

network_d:
  type: multiscale
  num_in_ch: 3
  num_feat: 64
  num_D: 2