Config everything in mcode/config.py
What to config?
- Model:
- Follow
mmseg
config pretrained
: path to ImageNet pretrained MiT backbone- Please change
pretrained
inbackbone
topretrained=pretrained
- Config model head to head of your choice
- Follow
- Wandb:
use_wandb
: True, False if debugwandb_key
: Please use your wandb authorize keywandb_name
: Name of your experiments, please make it as specific as possiblewandb_group
: We need 5 runs/experiments, grouping makes it easier to see on wandb UI
- Dataset:
train_images
: path to image in Train Datasettrain_masks
: path to mask in Train Datasettest_folder
: path to Test Datasettest_images
andtest_masks
: leave itnum_workers
: torch workerssave_path
: path to save checkpoints and logsbs
: this should be 16 if possiblegrad_accumulate_rate
: num iters to backward, ifbs=16
, this should be1