def main():
maybe_install_wordnet()
parser = get_parser()
args = parser.parse_args()
generate_hierarchy(**vars(args))
test_hierarchy(args)
generate_hierarchy_vis(args)
def NBDTLoss(cfg, criterion):
maybe_install_wordnet()
class_criterion = getattr(nbdtloss,
cfg.MODEL.ROI_RELATION_HEAD.LOSS.NBDT.TYPE)
NBDTloss = class_criterion(
dataset='VG150',
criterion=criterion,
sample_nums=cfg.MODEL.ROI_RELATION_HEAD.REL_SAMPLES[1:],
path_graph=cfg.MODEL.ROI_RELATION_HEAD.LOSS.NBDT.PATH_GRAPH,
path_wnids=cfg.MODEL.ROI_RELATION_HEAD.LOSS.NBDT.PATH_WNIDS,
tree_supervision_weight=cfg.MODEL.ROI_RELATION_HEAD.LOSS.NBDT.FACTOR)
return NBDTloss
import torch.optim as optim
import torch.nn.functional as F
import torch.backends.cudnn as cudnn
from nbdt import data, analysis, loss, models
import torchvision
import torchvision.transforms as transforms
import os
import argparse
import numpy as np
from nbdt.utils import (progress_bar, generate_fname, generate_kwargs, Colors,
maybe_install_wordnet)
maybe_install_wordnet()
datasets = ('CIFAR10', 'CIFAR100') + data.imagenet.names + data.custom.names
parser = argparse.ArgumentParser(description='PyTorch CIFAR Training')
parser.add_argument('--batch-size',
default=512,
type=int,
help='Batch size used for training')
parser.add_argument('--epochs',
'-e',
default=200,
type=int,
help='By default, lr schedule is scaled accordingly')
parser.add_argument('--dataset', default='CIFAR10', choices=datasets)
parser.add_argument('--arch',