| experiment_name = "unet_global_padding_nov_4" | |
| base_dir = "/exports/csce/eddie/eng/groups/DunnGroup/matthew/models_gelgenie" | |
| [processing] | |
| base_hardware = "EDDIE" | |
| device = "GPU" | |
| pe = 1 | |
| memory = 64 | |
| [data] | |
| n_channels = 1 | |
| batch_size = 2 | |
| num_workers = 1 | |
| val_percent = 10 | |
| dir_train_mask = [ "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/lsdb_gels/masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/nathan_gels/masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels/masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels_2/masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/neb_ladders/masks",] | |
| dir_train_img = [ "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/lsdb_gels/images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/nathan_gels/images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels/images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels_2/images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/neb_ladders/images",] | |
| dir_val_img = [ "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/lsdb_gels/val_images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/nathan_gels/val_images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels/val_images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels_2/val_images", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/neb_ladders/val_images",] | |
| dir_val_mask = [ "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/lsdb_gels/val_masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/nathan_gels/val_masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels/val_masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/matthew_gels_2/val_masks", "/exports/csce/eddie/eng/groups/DunnGroup/matthew/gel_data/neb_ladders/val_masks",] | |
| split_training_dataset = false | |
| apply_augmentations = true | |
| padding = true | |
| individual_padding = false | |
| weak_augmentations = false | |
| [model] | |
| model_name = "smp_unet" | |
| classes = 2 | |
| in_channels = 1 | |
| encoder_name = "resnet18" | |
| [training] | |
| loss = [ "dice", "crossentropy",] | |
| loss_component_weighting = [ 1, 1,] | |
| class_loss_weighting = false | |
| class_loss_weight_damper = [ 1.0, 1.0,] | |
| lr = 0.0001 | |
| epochs = 600 | |
| grad_scaler = true | |
| load_checkpoint = false | |
| optimizer_type = "adam" | |
| scheduler_type = "CosineAnnealingWarmRestarts" | |
| save_checkpoint = true | |
| checkpoint_frequency = 1 | |
| wandb_track = true | |
| model_cleanup_frequency = 20 | |
| wandb_id = "2ogdvood" | |
| [training.scheduler_specs] | |
| restart_period = 100 | |