Arguments

Description of arguments

Argument	Type	Description
Fix arguments
--author_name	str	author name
--directory_path, -dp	str	path to PRL2018_WIN
--dataset_name, -dn	str	dataset name
--img_size, -im	int	image size
--image_type, -it	str	image type includes BIN, RGB
--selection_mode, -sm	str	selection mode such as SAME or DIFF
Configuring arguments
--num_writers, -nw	int	number of writers
--num_training_patterns, -ntp	int	number of training patterns per writer
--local_feature, -lf	str	type of local feature type: subimg
--agg_mode, -am	str	different aggregation modes: average, max, kmax
--kmax, -k	int	value of k in kmax aggregation mode
--n_tuple, -nt	int	tuple size
--train_num_permutations, -trnp	int	number of permutations per epoch during training
--valid_num_permutations, -vanp	int	number of permutations per epoch during validating
--test_num_permutations, -tenp	int	number of permutations per epoch during testing
--eval_num_permutations', -evnp	int	number of permutations per epoch during evaluating
Training arguments
--lr, -l	float	learning rate
--max_epochs, -me	int	maximum number of epochs for training
--max_no_best, -mnb	int	early stop if accuracy does not increase during <max_no_best> epochs
--writer_per_batch, -wpb	int	number of writer per minibatch
--gpu, -g	int	gpu_id is used
--gpu_mem_ratio, -gmr	float	memory ratio of gpu from 0.0 to 1.0
Evaluation/Training arguments
--training, -t	bool	if evaluate model using False, else using default (True) for training
--resume, -r	str	if resume training process, pass the model filename
--global_step_start, -gss	int	if resume training process, pass the model global_step to continue training
--model_name, -mn	str	if evaluate model, pass the model filename.
--global_step_eval, -gse	int	if evaluate model, pass the model global_step to evaluate
--use_valid_data, -uvd	bool	use valid data or not while training model
--use_test_data, -utd	bool	use test data or not while training model
--eval_test_data, -etd	bool	use test data or not while evaluating model