Training

For pretraining the backbone edit simclr.yaml Mainly update the train_path

train_path: "/PATH/TO/DATASET/images" #remember to link only the images folder since this is no label training

Then run:

python3 main_pretrain.py \
    # path to training script folder
    --config-path scripts/pretrain/custom/ \
    # training config name
    --config-name simclr.yaml
    # add new arguments (e.g. those not defined in the yaml files)
    # by doing ++new_argument=VALUE
    # pytorch lightning's arguments can be added here as well.

Trained checkpoint for DeeplabV3+ backbone

The trained backbone checkpoint will be stored in folder trained_models in .ckpt format. Copy this path to add for loading checkpoint in DeeplabV3 backbone.