timelens_option.yml

# general settings
name: EvLowLight
model_type: EventVideoRecurrentTestModel
#scale: 2
num_gpu: 1  # official: 8 GPUs
manual_seed: 0
find_unused_parameters: true

# dataset and data loader settings
datasets:
  val:
    name: TimeLens
    type: EventVideoRecurrentTestDataset

    data_keys: ['ev', 'lq', 'gt' ]  #TODO: 'ev', 'lq', 'gt'
    ev_file_ext: .npy
    center_frame_only: true   #TODO:  [false], true
    filename_tmpl_ev: save.txt  #TODO: save.txt, events.h5, output.h5
    meta_info_file: data/meta_info/meta_info_real_timelens.txt    #TODO: meta info
    load_size: [ 640, 480 ]  #TODO: (854, 480)  (480, 320) (1280, 924)
#    real_ev_size: [305, 250]  # (320, 240), (346, 260), (960, 540)

    dataroot_gt: ../datasets/timelens/low
    dataroot_lq: ../datasets/timelens/low
    dataroot_ev: ../datasets/timelens/voxels

    scale: 1  #TODO:  2, 1
    filename_tmpl: 06d
    filename_ext: png

    num_frame: 5  # -1 not needed
    padding: reflection
    minimum_size: 16  #  8 * scale

# network structures
network_g:
  type: EvLowLightNet
  mid_channels: 64
  num_blocks: 7
  ev_flow_factor: 0.2
  prop_iter: 1
  flow_iter: 6

# path
path:
  pretrain_network_g: experiments/pretrained_models/EvLowLightNet.pth
  strict_load_g: false

# validation settings
val:
  save_img: true   # false
  suffix: null
  metrics: ~