RIDCP.yml

# general settings
name: vq_weight_dehaze_trained_on_ours
model_type: VQDehazeModel 
scale: &upscale 1
num_gpu: 4 # set num_gpu: 0 for cpu mode
manual_seed: 0

# dataset and data loader settings
datasets:
  train:
    name: General_Image_Train
    type: HazeOnlineDataset  
    dataroot_gt: datasets/rgb_500
    dataroot_depth: datasets/depth_500
    beta_range: [0.3, 1.5]
    A_range: [0.25, 1.0]
    color_p: 1.0
    color_range: [-0.025, 0.025]
    io_backend:
      type: disk
 
    gt_size: 256
    use_resize_crop: true
    use_flip: true
    use_rot: false

    # data loader
    use_shuffle: true
    batch_size_per_gpu: &bsz 4 
    num_worker_per_gpu: *bsz 
    dataset_enlarge_ratio: 1

    prefetch_mode: cpu
    num_prefetch_queue: *bsz

  val:
    name: General_Image_Train
    type: HazeOnlineDataset  
    dataroot_gt: datasets/rgb_500
    dataroot_depth: datasets/depth_500
    beta_range: [0.3, 1.5]
    A_range: [0.25, 1.0]
    color_p: 1.0
    color_range: [-0.025, 0.025]
    io_backend:
      type: disk

# network structures
network_g:
  type: VQWeightDehazeNet 
  gt_resolution: 256
  norm_type: 'gn'
  act_type: 'silu'
  scale_factor: *upscale
  codebook_params:
    - [64, 1024, 512]

  LQ_stage: true
  use_weight: false
  weight_alpha: -1.0
  frozen_module_keywords: ['quantize', 'decoder_group', 'after_quant_group', 'out_conv']

network_d:
  type: UNetDiscriminatorSN 
  num_in_ch: 512

# path
path:
  pretrain_network_hq: pretrained_models/pretrained_HQPs.pth
  pretrain_network_g: 
  pretrain_network_d: ~
  strict_load: false 
  # resume_state: ~

# training settings
train:
  optim_g:
    type: Adam
    lr: !!float 1e-4
    weight_decay: 0
    betas: [0.9, 0.99]
  optim_d:
    type: Adam
    lr: !!float 4e-4
    weight_decay: 0
    betas: [0.9, 0.99]

  scheduler:
    type: MultiStepLR
    milestones: [5000, 10000, 15000, 20000, 250000, 300000, 350000]
    gamma: 1 

  total_iter: 45000
  warmup_iter: -1  # no warm up

  # losses
  pixel_opt:
    type: L1Loss
    loss_weight: 1.0 
    reduction: mean

  perceptual_opt:
    type: LPIPSLoss
    loss_weight: !!float 1.0 
    
  gan_opt:
    type: GANLoss
    gan_type: hinge
    real_label_val: 1.0
    fake_label_val: 0.0
    loss_weight: 0.1

  codebook_opt:
    loss_weight: 1.0
  
  semantic_opt:
    loss_weight: 0.1 


  net_d_iters: 1
  net_d_init_iters: !!float 0 

# validation settings·
val:
  val_freq: !!float 80000
  save_img: true

  key_metric: psnr
  metrics:
    psnr: # metric name, can be arbitrary
      type: psnr 
      crop_border: 4
      test_y_channel: true
    ssim:
      type: ssim 
      crop_border: 4
      test_y_channel: true
    lpips:
      type: lpips 
      better: lower

# logging settings
logger:
  print_freq: 10
  save_checkpoint_freq: !!float 1e3
  save_latest_freq: !!float 5e2
  show_tf_imgs_freq: !!float 1e2
  use_tb_logger: true