Blind-Face-Super-Resolution/train_SRResNet_x4_FFHQ_300k.yml at main · anamikakolady/Blind-Face-Super-Resolution · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
# Modified SRResNet w/o BN from:
# Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

# ----------- Commands for running
# ----------- Single GPU with auto_resume
# CUDA_VISIBLE_DEVICES=0 python realesrgan/train.py -opt options/train_SRResNet_x4_FFHQ_300k.yml --auto_resume

# general settings
name: train_SRResNet_x4_FFHQ_300k
model_type: RealESRNetModel
scale: 4
num_gpu: 1  # set num_gpu: 0 for cpu mode
manual_seed: 0

# ----------------- options for synthesizing training data in RealESRNetModel ----------------- #
gt_usm: True  # USM the ground-truth
l1_gt_usm: true # use gt_usm as target for L1 loss

# the first degradation process
resize_prob: [0.2, 0.7, 0.1]  # up, down, keep
resize_range: [0.2, 1.5]
gaussian_noise_prob: 0.5
noise_range: [1, 20]
poisson_scale_range: [0.05, 2]
gray_noise_prob: 0.4
jpeg_range: [50, 95]

# the second degradation process
second_blur_prob: 0.8
resize_prob2: [0.3, 0.4, 0.3]  # up, down, keep
resize_range2: [0.3, 1.2]
gaussian_noise_prob2: 0.5
noise_range2: [1, 15]
poisson_scale_range2: [0.05, 1.5]
gray_noise_prob2: 0.4
jpeg_range2: [70, 95]

gt_size: 512
queue_size: 192  # divisible by batch size 24

# dataset and data loader settings
datasets:
  train:
    name: FFHQ_train
    type: FFHQsubDataset
    dataroot_gt: ../2024-S2-AI6126-Project2-Release/datasets/train/GT
    meta_info: ../2024-S2-AI6126-Project2-Release/datasets/train/meta_info_FFHQ6000sub_GT.txt
    io_backend:
      type: disk

    blur_kernel_size: 21
    kernel_list: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
    kernel_prob: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
    sinc_prob: 0.1
    blur_sigma: [0.2, 3]
    betag_range: [0.5, 4]
    betap_range: [1, 2]

    blur_kernel_size2: 21
    kernel_list2: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
    kernel_prob2: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
    sinc_prob2: 0.1
    blur_sigma2: [0.2, 1.5]
    betag_range2: [0.5, 4]
    betap_range2: [1, 2]

    final_sinc_prob: 0.8

    gt_size: 512
    use_hflip: True
    use_rot: False

    # data loader
    use_shuffle: true
    num_worker_per_gpu: 10  # earlier 6
    batch_size_per_gpu: 24 # earlier 8
    dataset_enlarge_ratio: 1
    prefetch_mode: ~

  val:
    name: FFHQ_val
    type: PairedImageDataset
    dataroot_gt: ../2024-S2-AI6126-Project2-Release/datasets/val/GT
    dataroot_lq: ../2024-S2-AI6126-Project2-Release/datasets/val/LQ
    io_backend:
      type: disk

# network structures
network_g:
  type: MSRResNet
  num_in_ch: 3
  num_out_ch: 3
  num_feat: 64
  num_block: 26
  upscale: 4

# path
path:
  pretrain_network_g: ~
  param_key_g: params
  strict_load_g: true
  resume_state: ~

# training settings
train:
  ema_decay: 0.999
  optim_g:
    type: Adam
    lr: !!float 1e-4
    weight_decay: 0
    betas: [0.9, 0.99]

  scheduler:
    type: CosineAnnealingRestartLR
    periods: [30000, 30000, 30000, 30000, 30000]
    restart_weights: [1, 1, 1, 1, 1]
    eta_min: !!float 1e-7

  total_iter: 150000 # 600 epochs
  warmup_iter: -1  # no warm up

  pixel_opt:
    type: MSELoss
    loss_weight: 1.0
    reduction: mean

  net_d_iters: 1
  net_d_init_iters: 0

# validation settings
val:
  val_freq: !!float 5e2
  save_img: true

  metrics:
    psnr: # metric name, can be arbitrary
      type: calculate_psnr
      crop_border: 0
      test_y_channel: false
      better: higher  # the higher, the better. Default: higher
    niqe:
      type: calculate_ssim
      crop_border: 0
      better: higher  # the lower, the better

# logging settings
logger:
  print_freq: 100
  save_checkpoint_freq: !!float 5e3
  use_tb_logger: true
  wandb:
    project: ~
    resume_id: ~

# dist training settings
dist_params:
  backend: nccl
  port: 29500