stdf-pytorch/option_R3_vimeo90k_2G.yml at master · ryanxingql/stdf-pytorch · 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
dataset:
  root: /media/x/Database/Vimeo-90K/

  train:  # LMDB
    type: Vimeo90KDataset

    # for lmdb
    gt_folder: vimeo_septuplet_ycbcr/
    lq_folder: vimeo_septuplet_ycbcr_intra/qp37/

    # for dataset
    gt_path: vimeo90k_train_gt.lmdb
    lq_path: vimeo90k_train_lq.lmdb
    meta_path: vimeo_septuplet/sep_trainlist.txt
    gt_size: 128  # ground truth patch size: gt_size * gt_size
    use_flip: True
    use_rot: True  # rotation per 90 degrees
    random_reverse: False

    # for datasampler
    enlarge_ratio: 300  # enlarge dataset by randomly cropping.

    # for dataloader
    num_worker_per_gpu: 6  # 12 in total. mainly affect IO speed
    batch_size_per_gpu: 16  # bs=32, divided by 2 GPUs

  val:  # Disk IO
    type: VideoTestVimeo90KDataset
    gt_path: vimeo_septuplet_ycbcr/
    lq_path: vimeo_septuplet_ycbcr_intra/qp37/
    meta_path: vimeo_septuplet/sep_testlist.txt
    #enlarge_ratio: 1
    #use_flip: False
    #use_rot: False
    #random_reverse: False

  test:
    type: VideoTestVimeo90KDataset
    gt_path: vimeo_septuplet_ycbcr/
    lq_path: vimeo_septuplet_ycbcr_intra/qp37/
    meta_path: vimeo_septuplet/sep_testlist.txt

network:
  radius: 3  # total num of input frame = 2 * radius + 1

  stdf:
    in_nc: 1  # 1 for Y
    out_nc: 64
    nf: 32  # num of feature maps
    nb: 3  # num of conv layers
    base_ks: 3
    deform_ks: 3  # size of the deformable kernel

  qenet:
    in_nc: 64  # = out_nc of stdf
    out_nc: 1  # 1 for Y
    nf: 48
    nb: 8
    base_ks: 3

train:
  exp_name: Vimeo90K_R3_enlarge300x  # default: timestr. None: ~
  random_seed: 7
  pre-val: True  # evaluate criterion before training, e.g., ori PSNR
  num_iter: !!float 3e+5
  interval_print: !!float 100
  interval_val: !!float 5e+3  # also save model
  pbar_len: 100

  optim:
    type: Adam
    lr: !!float 1e-4  # init lr of scheduler
    betas: [0.9, 0.999]
    eps: !!float 1e-08

  scheduler:
    is_on: False
    type: CosineAnnealingRestartLR
    periods: [!!float 5e+4, !!float 5e+4, !!float 5e+4, !!float 5e+4, !!float 5e+4, !!float 5e+4]  # epoch interval
    restart_weights: [1, 0.5, 0.5, 0.5, 0.5, 0.5]
    eta_min: !!float 1e-7

  loss:
    type: CharbonnierLoss
    eps: !!float 1e-6

  criterion:
    type: PSNR
    unit: dB

test:
  restore_iter: !!float 610000
  pbar_len: 100

  criterion:
    type: PSNR
    unit: dB