print('gradients --- ', k)
if cfg.solver == 'SGD':
optimizer = torch.optim.SGD()
elif cfg.solver == 'RMS':
optimizer = torch.optim.RMSprop()
elif cfg.solver == 'Adam':
optimizer = torch.optim.Adam()
else:
optimizer = ''
raise ValueError()
model.cuda()
# # DATA LOADER
get_loader = get_data_loader(cfg.datasetname)
train_data = get_loader()
class_names = train_data.dataset.classes
print('dataset len: {}'.format(len(train_data.dataset)))
tb_dir = os.path.join(cfg.train_dir, cfg.backbone + '_' + cfg.datasetname,
time.strftime("%h%d_%H"))
writer = tbx.FileWriter(tb_dir)
summary_out = []
global_step = 0
timer = Timer()
for ep in range(start_epoch, cfg.max_epoch):
if ep in cfg.lr_decay_epoches and cfg.solver == 'SGD':
lr *= cfg.lr_decay
def parse_args():
parser = argparse.ArgumentParser(description='')
parser.add_argument('--json',
dest='json',
help='citerpersons json file',
default='',
type=str)
args = parser.parse_args()
if len(sys.argv) == 1:
parser.print_help()
sys.exit(1)
return args
args = parse_args()
if __name__ == '__main__':
get_loader = get_data_loader('citypersons')
test_data = get_loader('./data/citypersons',
'val',
is_training=False,
batch_size=1,
num_workers=1,
shuffle=False)
dataset = test_data.dataset
if not os.path.exists(args.json):
print('%s do not exist' % args.json)
dataset.eval_over_scales(args.json)