deep-supervised-contrastive

Research Code for deepfake & gan classification.

train.py usage:

usage: train.py [-h] -p project_name [-rg RUN_GROUP] [--name name] [-e epochs] [-b batch_size] [-opt optimizer] [-lr learning_rate] [-wd weight_decay] [-sch scheduler] [-step scheduler_step_size] [-gamma scheduler_gamma] -i
                input_size [-cc crop_size] -am augmentations_mode -at augmentations_type -d dataset -dp dataset_path [-trainp train_path] [-valp validation_path] [-testp test_path] [-proj projector [projector ...]]
                [-savem save_model_path] [-saveb save_backbone_path] -dev device [-nw num_workers] [-fp]

Training arguments

optional arguments:
  -h, --help            show this help message and exit
  -p project_name, --project_name project_name
                        Project name, utilized for logging purposes in W&B.
  -rg RUN_GROUP, --run-group RUN_GROUP
                        group of runs to put the current run into (e.g. ff)
  --name name           Experiment name that logs into wandb.
  -e epochs, --epochs epochs
                        Max number of epochs to train for
  -b batch_size, --batch_size batch_size
                        Input batch size for training (default: 32).
  -opt optimizer, --optimizer optimizer
                        optimizer to use during training (default: adam).
  -lr learning_rate, --learning_rate learning_rate
                        Learning rate of the optimizer (default: 1e-3).
  -wd weight_decay, --weight_decay weight_decay
                        Weight decay of the optimizer (default: 1e-5).
  -sch scheduler, --scheduler scheduler
                        Scheduler to use during training (default: steprl).
  -step scheduler_step_size, --scheduler_step_size scheduler_step_size
                        scheduler step size (default: 5)
  -gamma scheduler_gamma, --scheduler_gamma scheduler_gamma
                        scheduler gamma (default: 0.1)
  -i input_size, --input_size input_size
                        input size for models
  -cc crop_size, --crop_size crop_size
                        crop size for models
  -am augmentations_mode, --augmentations_mode augmentations_mode
                        augmentations mode for transforms (gan or df)
  -at augmentations_type, --augmentations_type augmentations_type
                        augmentations type for the dataset
  -d dataset, --dataset dataset
                        dataset name on which to evaluate
  -dp dataset_path, --dataset_path dataset_path
                        root dataset path on which to evaluate
  -trainp train_path, --train_path train_path
                        Training dataset path for csv.
  -valp validation_path, --validation_path validation_path
                        Validation dataset path for csv.
  -testp test_path, --test_path test_path
                        test dataset path for csv.
  -proj projector [projector ...], --projector projector [projector ...]
                        projector architecture
  -savem save_model_path, --save_model_path save_model_path
                        Save directory path for model.
  -saveb save_backbone_path, --save_backbone_path save_backbone_path
                        Save directory path for backbone net.
  -dev device, --device device
                        Device used during training
  -nw num_workers, --num-workers num_workers
                        number of workers to use for dataloading (default: 8)
  -fp, --fp16           boolean for using mixed precision.```