Log Details

Logs

• Training and testing logs will be saved under [LOG_ROOT]/PG2022_RealTime_VDBLR/[mode]/:

  [LOG_ROOT]
   └── PG2022_RealTime_VDBLR
       ├── [mode]
       │   ├── checkpoint      # model checkpoints and resume states
       │   ├── log             # scalar/image logs for tensorboard
       │   ├── sample          # sample images of training and validation
       │   ├── config          # config file
       │   ├── result          # resulting images of evaluation
       │   ├── cost.txt        # network size and MACs measured on an image with the size of (1, 3, 1280, 720)
       │   └── [network].py    # network file
       └── ...
  

• In ./config/config.py, you may configure the following items:

  • config.log_offset: configures [LOG_ROOT]. Default: ./logs
  • config.write_ckpt_every_epoch: configures an epoch period for saving checkpoints, resume states and scalar logs. Default: 4
  • config.write_log_every_itr: configures an iteration period for saving sample images. Default: {'train':65, 'valid':20}
  • config.refresh_image_log_every_epoch: configures an epoch period for erasing sample images. Default: {'train':16, 'valid': 16}