test_vsrTransformer_x4_REDS.yml

name: test_vsrTransformer_x4_reds
model_type: EDVRModel
scale: 4
num_gpu: 4  # set num_gpu: 0 for cpu mode
manual_seed: 0
crop: true

datasets:
  test:
    name: REDS4
    type: VideoTestDataset
    dataroot_gt: /cluster/work/cvl/videosr/REDS/train_sharp
    dataroot_lq: /cluster/work/cvl/videosr/REDS/train_sharp_bicubic/X4/
    meta_info_file: basicsr/data/meta_info/meta_info_REDS4_test_GT.txt
    io_backend:
      type: disk

    all_gt:  false # True #  TODO

    cache_data: false
    num_frame: 5
    padding: reflection_circle  #TODO

    lq_size: 64

# network structures
network_g:
  type: vsrTransformer
  image_ch: 3
  num_feat: 64 
  num_frame: 5  
  num_extract_block: 5 
  num_reconstruct_block: 30 
  depth: 5  
  heads: 1   
  patch_size: 8  
  spynet_pretrained: '/cluster/work/cvl/videosr/pretrained_models/spynet/spynet.pth'   

# path
path:
  pretrain_network_g: /cluster/work/cvl/videosr/experiments/net_g_latest.pth 
  strict_load_g: true
  root: /cluster/work/cvl/videosr/experiments/

# validation settings
val:
  save_img: False #True #
  suffix: ~  # add suffix to saved images, if None, use exp name

  metrics:
    psnr: # metric name, can be arbitrary
      type: calculate_psnr # calculate_ssim #
      crop_border: 0
      test_y_channel: False #True