# python train.py -opt options/sr/x1_ITF_SkinDiffDetail_Lite_v1.yml
name: x1_ITF_SkinDiffDetail_Lite_v1
# the name that defines the experiment and the directory that will be created in the experiments directory.
# name: debug_001_template  # use the "debug" or "debug_nochkp" prefix in the name to run a test session and check everything is working. Does validation and state saving every 8 iterations. Remove "debug" to run the real training session.
use_tb_logger: false
# wheter to enable Tensorboard logging or not. Output will be saved in: traiNNer/tb_logger/
model: sr 
# the model training strategy to be used. Depends on the type of model, from: https://github.com/victorca25/traiNNer/tree/master/codes/models
scale: 1 # the scale factor that will be used for training for super-resolution cases. Default is "1".
gpu_ids: [0] # the list of `CUDA_VISIBLE_DEVICES` that will be used during training, ie. for two GPUs, use [0, 1]. The batch size should be a multiple of the number of 'gpu_ids', since images will be distributed from the batch to each GPU.
use_amp: true # select to use PyTorch's Automatic Mixed Precision package to train in low-precision FP16 mode (lowers VRAM requirements).
use_swa: false # select to use Stochastic Weight Averaging
use_cem: false # select to use CEM during training. https://github.com/victorca25/traiNNer/tree/master/codes/models/modules/architectures/CEM

# Dataset options:
datasets: # configure the datasets
  train: # the stage the dataset will be used for (training)
    name: x1_ITF_SkinDiffDetail_Lite_v1 # the name of your dataset (only informative)
    mode: aligned 
    # dataset mode: https://github.com/victorca25/traiNNer/tree/master/codes/data
    dataroot_HR: [
      #'K:/TRAINING/data/Skin_Diff2Nrml/hr_clean_tiles/'
      '../datasets/Skin_DiffDetail/hr/'
      ]
    dataroot_LR: [
      #'K:/TRAINING/data/Skin_Diff2Nrml/lr_clean_tiles/'
      '../datasets/Skin_DiffDetail/lr_soft/'
      ] # low resolution images
    subset_file: null
    use_shuffle: true
    znorm: false
    n_workers: 8
    batch_size: 12
    virtual_batch_size: 12
    preprocess: crop
    crop_size: 64
    image_channels: 3

    # Color space conversion
    # color: 'y'
    # color_LR: 'y'
    # color_HR: 'y'
    
    # LR and HR modifiers.
    # aug_downscale: 0.2
    # shape_change: reshape_lr
    
    # Enable random downscaling of HR images (will fix LR pair to correct size)
    hr_downscale: true
    hr_downscale_types: [0, 3]
    hr_downscale_amount: [1, 2, 4]
    # #pre_crop: true
    
    # Presets and on the fly (OTF) augmentations
    #augs_strategy: combo
    #add_blur_preset: custom_blur
    #add_resize_preset: custom_resize
    #add_noise_preset: custom_noise
    #aug_downscale: 0.2
    resize_strat: pre
    
    # On the fly generation of LR:
    # dataroot_kernels: 'KERNEL PATH !!!! CHANGE THIS OR COMMENT OUT'
    #lr_downscale: false
    #lr_downscale_types: ["linear", "bicubic", "nearest_aligned"]

    # Rotations augmentations:
    use_flip: true
    use_rot: true
    use_hrrot: true
    
    # Noise and blur augmentations:
    #lr_blur: true
    #lr_blur_types: {sinc: 0.2, iso: 0.2, ansio2: 0.4, sinc2: 0.2, clean: 3}
    #noise_data: 'K:/TRAINING/traiNNer/noise_patches/'
    #lr_noise: true
    #lr_noise_types: {camera: 0.1, jpeg: 0.8, clean: 3}
    #lr_noise2: false
    #lr_noise_types2: {jpeg: 1, webp: 0, clean: 2, camera: 2}
    #hr_noise: false
    #hr_noise_types:  {gaussian: 1, clean: 4}
    
    # Color augmentations
    # lr_fringes: false
    # lr_fringes_chance: 0.4
    # auto_levels: HR
    # rand_auto_levels: 0.7
    #lr_unsharp_mask: true
    #lr_rand_unsharp: 0.7
    # hr_unsharp_mask: true
    # hr_rand_unsharp: 1
    
    # Augmentations for classification or (maybe) inpainting networks:
    # lr_cutout: false
    # lr_erasing: false

  #val: 
    #name: val_set14_part
    #mode: aligned
    #dataroot_B: '../datasets/val/hr'
    #dataroot_A: '../datasets/val/lr'
    
    #znorm: false
    
    # Color space conversion:
    # color: 'y'
    # color_LR: 'y'
    # color_HR: 'y'
    

path:
    root: '../'
    pretrain_model_G: '../experiments/pretrained_models/1x_DIV2K-Lite_SpongeBC1-Lite_interp.pth'
    # pretrain_model_D: 'K:/TRAINING/data/models/x1_ITF_SkinDiff2Nrm_Lite_v3_208500_D.pth'
    resume_state: '../experiments/x1_ITF_SkinDiffDetail_Lite_v1/training_state/latest.state'

# Generator options:
network_G: esrgan-lite # configurations for the Generator network


# Discriminator options:
network_D:
    # ESRGAN (default)| PPON:
    which_model_D: multiscale # discriminator_vgg_128 | discriminator_vgg | discriminator_vgg_128_fea (feature extraction) | patchgan | multiscale
    norm_type: batch
    act_type: leakyrelu
    mode: CNA # CNA | NAC
    nf: 32
    in_nc: 3
    nlayer: 3 # only for patchgan and multiscale
    num_D: 3 # only for multiscale

train:
    # Optimizer options:
    optim_G: adamp
    optim_D: adamp
    
    # Schedulers options:
    lr_scheme: MultiStepLR
    lr_steps_rel: [50000, 100000, 200000, 300000]
    lr_gamma: 0.5

    # For SWA scheduler
    swa_start_iter_rel: 0.05
    swa_lr: 1e-4
    swa_anneal_epochs: 10
    swa_anneal_strategy: "cos"
    
    # Losses:
    pixel_criterion: l1  # pixel (content) loss
    pixel_weight: 0.05
    feature_criterion: l1 # feature loss (VGG feature network)
    feature_weight: 0.3
    cx_type: contextual  # contextual loss
    cx_weight: 1
    cx_vgg_layers: {conv_3_2: 1, conv_4_2: 1}
    #hfen_criterion: l1  # hfen
    #hfen_weight: 1e-6 
    #grad_type: grad-4d-l1  # image gradient loss
    #grad_weight: 4e-1
    # tv_type: normal  # total variation
    # tv_weight: 1e-5
    # tv_norm: 1
    ssim_type: ssim  # structural similarity
    ssim_weight: 0.05
    lpips_weight: 0.25  # [.25] perceptual loss
    lpips_type: net-lin
    lpips_net: squeeze
    
    # Experimental losses
    # spl_type: spl  # spatial profile loss
    # spl_weight: 0.1
    #of_type: overflow  # overflow loss
    #of_weight: 0.1
    # range_weight: 1  # range loss
    # fft_type: fft  # FFT loss
    # fft_weight: 0.2 #[.2]
    color_criterion: color-l1cosinesim  # color consistency loss
    color_weight: 0.1
    # avg_criterion: avg-l1  # averaging downscale loss
    # avg_weight: 5
    # ms_criterion: multiscale-l1  # multi-scale pixel loss
    # ms_weight: 1e-2
    #fdpl_type: fdpl  # frequency domain-based perceptual loss
    #fdpl_weight: 1e-3
    
    # Adversarial loss:
    #gan_type: vanilla
    #gan_weight: 4e-3
    # freeze_loc: 4
    # For wgan-gp:
    # D_update_ratio: 1
    # D_init_iters: 0
    # gp_weigth: 10
    # Feature matching (if using the discriminator_vgg_128_fea or discriminator_vgg_fea):
    # gan_featmaps: true
    # dis_feature_criterion: cb  # discriminator feature loss
    # dis_feature_weight: 0.01
    
    # For PPON:
    # p1_losses: [pix]
    # p2_losses: [pix-multiscale, ms-ssim]
    # p3_losses: [fea]
    # ppon_stages: [1000, 2000]
    
    # Differentiable Augmentation for Data-Efficient GAN Training
    # diffaug: true
    # dapolicy: 'color,transl_zoom,flip,rotate,cutout'
    
    # Batch (Mixup) augmentations
    #mixup: false
    #mixopts: [blend, rgb, mixup, cutmix, cutmixup] # , "cutout", "cutblur"]
    #mixprob: [1.0, 1.0, 1.0, 1.0, 1.0] #, 1.0, 1.0]
    #mixalpha: [0.6, 1.0, 1.2, 0.7, 0.7] #, 0.001, 0.7]
    #aux_mixprob: 1.0
    #aux_mixalpha: 1.2
   # mix_p: 1.2
    
    # Frequency Separator
    #fs: true
    #lpf_type: average
    #hpf_type: average
    
    # Other training options:
    manual_seed: 0
    niter: 250000
    # warmup_iter: -1
    #val_freq: 5e3
    # overwrite_val_imgs: true
    # val_comparison: true
    # metrics: 'psnr,ssim,lpips'
    #grad_clip: auto
    #grad_clip_value: 0.1 # "auto"

logger:
    print_freq: 50
    save_checkpoint_freq: 500
    overwrite_chkp: false