# do not remove
from localizer.kneel_before_wrist.model import HourglassNet_PTL
IND_TO_SIDE = {0: 'PA', 1: 'LAT'}
if __name__ == '__main__':
cwd = Path().cwd()
conf_file = cwd.parents[0] / 'config' / 'config.yaml'
config = get_conf(conf_file=conf_file, cwd=cwd)
# save config
os.makedirs(config.snapshot_dir, exist_ok=True)
OmegaConf.save(config=config,
f=os.path.join(config.snapshot_dir, 'params.yaml'))
apply_fixed_seed(config.seed)
apply_deterministic_computing(config.deterministic)
meta_path = os.path.join(config.dataset.data_home,
config.dataset.data_folder, config.dataset.meta)
# master copy of meta data
master_meta = pd.read_csv(meta_path)
# check the number of sides
if isinstance(config.dataset.side, int):
config.dataset.side = [config.dataset.side]
for side in config.dataset.side:
train_meta = master_meta[master_meta.Side == side]
landmark_trf = get_landmark_transform_kneel(config)
# get mean and std
ms_file = os.path.join(config.dataset.data_home,
edgecolor='k',
histtype='bar',
rwidth=0.9)
plt.xlabel('Entropy')
plt.ylabel('Count')
plt.legend()
plt.savefig(path, bbox_inches='tight')
plt.close()
if __name__ == '__main__':
# create arguments
cwd = Path().cwd()
conf_file = cwd.parents[0] / 'config' / 'config.yaml'
config = get_conf(conf_file=conf_file, cwd=cwd)
apply_fixed_seed(seed=config.seed)
apply_deterministic_computing(config.deterministic)
meta = pd.read_csv(config.dataset.meta)
pa = 0
lat = 1
if isinstance(config.local_rank, int):
device = torch.device(f'cuda:{config.local_rank}')
torch.cuda.set_device(config.local_rank)
else:
device = torch.device('cpu')
with open('temp_old.pkl', 'rb') as f:
temp_dict = pickle.load(f)
pa_temp = temp_dict['PA']
lat_temp = temp_dict['LAT']