test_drunet.json

{
  "task": "drunet"  //  root/task/images-models-options
  , "model": "plain" // "plain"
  , "gpu_ids": [0]

  , "scale": 0       // broadcast to "netG" if SISR
  , "n_channels": 1  // broadcast to "datasets", 1 for grayscale, 3 for color
  , "n_channels_datasetload": 3 // broadcast to image training set

  , "path": {
    "root": "drunet_inference"            // "denoising" | "superresolution"
    , "pretrained_netG": "path/to/model"    // path of pretrained model, if model from scratch type: null
  }

  , "datasets": {
    "train": {
      "name": "train_dataset"           // just name
      , "dataset_type": "drunet"         // "dncnn" | "dnpatch" for dncnn,  | "fdncnn" | "ffdnet" | "sr" | "srmd" | "dpsr" | "plain" | "plainpatch"
      , "dataroot_H": "web_subset/ground_truth" // path of H training dataset
      , "sigma": [0, 0]      // 15, 25, 50 for DnCNN | [0, 75] for FFDNet and FDnCNN
      , "num_patches_per_image": 20   // number of random patches of training image
      , "dataroot_L": "web_subset/capturas" // path of L training dataset, if using noisy H type: null
      , "H_size": 128                    // patch size 40 | 64 | 96 | 128 | 192
      , "dataloader_shuffle": false
      , "dataloader_num_workers": 8
      , "dataloader_batch_size": 64     // batch size 1 | 16 | 32 | 48 | 64 | 128
    }
    , "test": {
      "name": "test_dataset"            // just name
      , "dataset_type": "ffdnet"         // "dncnn" | "dnpatch" for dncnn,  | "fdncnn" | "ffdnet" | "sr" | "srmd" | "dpsr" | "plain" | "plainpatch"
      , "dataroot_H": "web_subset/capturas_splitter"  // path of H testing dataset
      , "dataroot_L": "web_subset/capturas_splitter"   // path of L testing dataset
      , "sigma_test": 0 // 15, 25, 50 for DnCNN and ffdnet
    }
  }

  , "netG": {
    "net_type": "drunet" // "dncnn" | "fdncnn" | "ffdnet" | "srmd" | "dpsr" | "srresnet0" |  "srresnet1" | "rrdbnet" 
    , "in_nc": 2        // input channel number
    , "out_nc": 1       // ouput channel number
    , "nc": [64, 128, 256, 512]          // 64 for "dncnn"
    , "nb": 4          // 17 for "dncnn", 20 for dncnn3, 16 for "srresnet"
    , "gc": 32          // unused
    , "ng": 2           // unused
    , "reduction": 16  // unused
    , "act_mode": "R"  // "BR" for BN+ReLU | "R" for ReLU
    , "upsample_mode": "convtranspose"  // "pixelshuffle" | "convtranspose" | "upconv"
    , "downsample_mode": "strideconv"   // "strideconv" | "avgpool" | "maxpool"
    , "bias": false//
    , "init_type": "orthogonal"         // "orthogonal" | "normal" | "uniform" | "xavier_normal" | "xavier_uniform" | "kaiming_normal" | "kaiming_uniform"
    , "init_bn_type": "uniform"         // "uniform" | "constant"
    , "init_gain": 0.2
  }

  , "train": {
      "epochs": 1000                      // number of epochs to train
    , "G_lossfn_type": "l2"               // "l1" preferred | "l2sum" | "l2" | "ssim" 
    , "G_lossfn_weight": 1.0            // default
    , "G_tvloss_weight": 0.1            // total variation weight

    , "G_optimizer_type": "adam"        // fixed, adam is enough
    , "G_optimizer_lr": 1e-4           // learning rate
    , "G_optimizer_clipgrad": null      // unused

    , "G_scheduler_type": "MultiStepLR" // "MultiStepLR" is enough
    , "G_scheduler_milestones": [1600,  3200,  4800,  6400,  8000,  9600, 11200, 12800, 14400]
    , "G_scheduler_gamma": 0.1  //

    , "G_regularizer_orthstep": null    // unused
    , "G_regularizer_clipstep": null    // unused
    
    // iteration (batch step) checkpoints
    , "checkpoint_test": 1600           // for testing
    , "checkpoint_save": 3999           // for saving model
    , "checkpoint_print": 16           // for print
  }
}