erase_wrong_plotting_summaries(args.folder,
list(args.validation_datasets))
if args.restart_validations:
erase_validations(args.folder, list(args.validation_datasets))
if args.single_process is not None:
if args.exp is None:
raise ValueError(
" You should set the exp alias when using single process")
create_exp_path(args.folder, args.exp)
if args.single_process == 'train':
# TODO make without position, increases the legibility.
execute_train("0", args.folder, args.exp, False)
elif args.single_process == 'validation':
if len(args.validation_datasets) == 1:
execute_validation("0", args.folder, args.exp,
args.validation_datasets[0], args.model,
False)
else:
execute_validation("0", args.folder, args.exp,
args.validation_datasets, args.model, False)
elif args.single_process == 'drive':
driving_environments = fix_driving_environments(
list(args.driving_environments))
execute_drive("0",
# There are two modes of execution
if args.single_process is not None:
####
# MODE 1: Single Process. Just execute a single experiment alias.
####
if args.exp is None:
raise ValueError(
" You should set the exp alias when using single process")
create_exp_path(args.folder, args.exp)
if args.single_process == 'train':
execute_train(gpu="0",
exp_batch=args.folder,
exp_alias=args.exp,
suppress_output=False,
number_of_workers=args.number_of_workers)
elif args.single_process == 'validation':
execute_validation(gpu="0",
exp_batch=args.folder,
exp_alias=args.exp,
dataset=args.validation_datasets[0],
suppress_output=False)
elif args.single_process == 'drive':
drive_params['suppress_output'] = False
execute_drive("0", args.folder, args.exp,
list(args.driving_environments)[0], drive_params)
try:
int(gpu)
except:
raise ValueError(" Gpu is not a valid int number")
# Obs this is like a fixed parameter, how much a validation and a train and drives ocupies
# TODO: MAKE SURE ALL DATASETS ARE " WAYPOINTED "
# execute_train("0", "eccv", "experiment_1")
# execute_validation("0", "eccv", "experiment_1", "SeqVal")
if args.single_process is not None:
if args.single_process == 'train':
# TODO make without position, increases the legibility.
execute_train("1", "eccv", "experiment_unit_task")
# execute_train("2", "eccv", "experiment_HighWeather12_nolane")
if args.single_process == 'validation':
execute_validation("0", "eccv", "experiment_1", "SeqVal")
# if args.single_process == 'drive':
# execute_drive("0", "eccv", "experiment_1", 'Town02')
# else:
# # TODO: of course this change from gpu to gpu , but for now we just assume at least a K40
# # Maybe the latest voltas will be underused
# # OBS: This usage is also based on my tensorflow experiences, maybe pytorch allows more.
# allocation_parameters = {'gpu_value': 3.5,
elif all(v is None for v in [
args.encoder_checkpoint, args.encoder_folder,
args.encoder_exp
]):
encoder_params = None
else:
print(args.encoder_folder, args.encoder_exp,
args.encoder_checkpoint)
raise ValueError(
"You should set all three arugments for using encoder: --encoder-folder, --encoder-exp and --encoder-checkpoint"
)
if args.single_process == 'train':
execute_train(gpu=args.gpus[0],
exp_batch=args.folder,
exp_alias=args.exp,
suppress_output=False,
encoder_params=encoder_params)
elif args.single_process == 'validation':
execute_validation(gpu=args.gpus[0],
exp_batch=args.folder,
exp_alias=args.exp,
json_file_path=args.val_json,
suppress_output=False,
encoder_params=encoder_params)
# train_encoder and validation_encoder are for training the encoder model only.
elif args.single_process == 'train_encoder':
# Check if the mandatory folder argument is passed
if args.encoder_folder is None:
raise ValueError(
from coil_core import execute_train
from coilutils.general import create_log_folder, create_exp_path, erase_logs
if __name__ == '__main__':
folder = 'cvpr'
exp = 'img_gtseg_camv_control'
create_log_folder(folder)
erase_logs(folder)
create_exp_path(folder, exp)
execute_train('0', folder, exp)
print("SUCCESSFULLY RAN TRAINING")