cvfinetune/examples/basic/utils/parser.py

import os

from cvargparse import GPUParser, Arg
from chainer_addons.links import PoolingType

from cvfinetune.parser import default_factory
from cvfinetune import parser as parser_module


def parse_args():

	parser = parser_module.FineTuneParser(default_factory([

			Arg("--pretrained_on", choices=["inat", "imagenet"],
				help="network pretraining"),


			# Arg("--normalize", action="store_true",
			# 	help="normalize features after cbil- or alpha-poolings"),

			# Arg("--subset", "-s", type=int, nargs="*", default=[-1], help="select specific classes"),
			# Arg("--no_sacred", action="store_true", help="do save outputs to sacred"),

			# Arg("--use_parts", action="store_true",
			# 	help="use parts, if present"),
			# Arg("--simple_parts", action="store_true",
			# 	help="use simple parts classifier, that only concatenates the features"),
			# Arg("--no_global", action="store_true",
			# 	help="use parts only, no global feature"),


			# Arg("--parts_in_bb", action="store_true", help="take only uniform regions where the centers are inside the bounding box"),

			# Arg("--rnd_select", action="store_true", help="hide random uniform regions of the image"),
			# Arg("--recurrent", action="store_true", help="observe all parts in recurrent manner instead of the whole image at once"),

			# ## AlphaPooling options
			# Arg("--init_alpha", type=int, default=1, help="initial parameter for alpha pooling"),
			# Arg("--kappa", type=float, default=1., help="Learning rate factor for alpha pooling"),
			# Arg("--switch_epochs", type=int, default=0, help="train alpha pooling layer and the rest of the network alternating")
		])
	)

	return parser.parse_args()