'--shared_transformer',
str(shared_transformer),
'--patch_center',
'mean',
'--training_order',
'random_shape_consecutive',
'--use_point_stn',
str(1),
'--uniform_subsample',
str(0),
'--outputs',
]
train_params += features
# train model on GT data with multiple query points per patch
train_opt = points_to_surf_train.parse_arguments(train_params)
points_to_surf_train.points_to_surf_train(train_opt)
valsets = [
'abc_minimal',
]
for valset in valsets:
# validate model
in_dir_val = os.path.join(base_dir, valset)
out_dir_val = os.path.join('results', model_name, valset)
res_dir_eval = os.path.join(out_dir_val, 'eval')
eval_params = [
'--indir',
in_dir_val,
'--outdir',
out_dir_val,
from source import points_to_surf_train
# When you see this error:
# 'Expected more than 1 value per channel when training...' which is raised by the BatchNorm1d layer
# for multi-gpu, use a batch size that can't be divided by the number of GPUs
# for single-gpu, use a straight batch size
# see https://github.com/pytorch/pytorch/issues/2584
# see https://forums.fast.ai/t/understanding-code-error-expected-more-than-1-value-per-channel-when-training/9257/12
def full_train(opt):
points_to_surf_train.points_to_surf_train(opt)
if __name__ == '__main__':
# train model on GT data with multiple query points per patch
full_train(opt=points_to_surf_train.parse_arguments())
print('MeshNet training is finished!')