print(checkpoint_dir)
if params.save_iter != -1:
modelfile = get_assigned_file(checkpoint_dir,params.save_iter)
else:
modelfile = get_best_file(checkpoint_dir)
if params.save_iter != -1:
outfile = os.path.join( checkpoint_dir.replace("checkpoints","features"), split + "_" + str(params.save_iter)+ ".hdf5")
else:
outfile = os.path.join( checkpoint_dir.replace("checkpoints","features"), split + ".hdf5")
datamgr = SimpleDataManager(image_size, batch_size = 3)
if params.dct_status:
data_loader = datamgr.get_data_loader_dct(loadfile, aug = False, filter_size = params.filter_size)
else:
data_loader = datamgr.get_data_loader(loadfile, aug = False)
# if params.method =='baseline++':
# model = BaselineTrain( model_dict[params.model], params.num_classes, loss_type = 'dist')
if params.method == 'S2M2_R':
if params.dataset == 'cifar':
model = wrn_mixup_model.wrn28_10(num_classes = params.num_classes, dct_status = params.dct_status)
else:
#model = wrn_mixup_model.wrn28_10(200)
model = wrn_mixup_model.wrn28_10(num_classes = 200, dct_status = params.dct_status)
else:
model = model_dict[params.model]()
if params.model == 'WideResNet28_10':
image_size = 84
params.num_classes = 200
else:
image_size = 224
params.num_classes = 200
print(params.checkpoint_dir)
start_epoch = params.start_epoch
stop_epoch = params.stop_epoch
if params.method in ['baseline++', 'S2M2_R', 'rotation']:
if params.dct_status:
base_datamgr = SimpleDataManager(image_size_dct,
batch_size=params.batch_size)
base_loader = base_datamgr.get_data_loader_dct(
base_file,
aug=params.train_aug,
filter_size=params.filter_size)
base_datamgr_test = SimpleDataManager(
image_size_dct, batch_size=params.test_batch_size)
base_loader_test = base_datamgr_test.get_data_loader_dct(
base_file, aug=False, filter_size=params.filter_size)
test_few_shot_params = dict(n_way=params.train_n_way,
n_support=params.n_shot)
val_datamgr = SetDataManager(image_size_dct,
n_query=15,
**test_few_shot_params)
val_loader = val_datamgr.get_data_loader_dct(
val_file, aug=False, filter_size=params.filter_size)
else:
base_datamgr = SimpleDataManager(image_size,
batch_size=params.batch_size)
loadfile = configs.data_dir[params.dataset] + 'novel.json'
if params.dct_status:
image_size = 56
else:
if params.dataset == 'cifar':
image_size = 32
params.num_classes = 64
else:
image_size = 84
#if params.dataset == 'miniImagenet' or params.dataset == 'CUB':
datamgr = SimpleDataManager(image_size, batch_size=16)
if params.dct_status:
novel_loader = datamgr.get_data_loader_dct(loadfile, aug=False)
else:
novel_loader = datamgr.get_data_loader(loadfile, aug=False)
params.channels = 3
checkpoint_dir = '%s/checkpoints/%s/%s_%s_%sway_%sshot' % (
configs.save_dir, params.dataset, params.model, params.method,
params.test_n_way, params.n_shot)
if params.train_aug:
checkpoint_dir += '_aug'
if params.dct_status:
checkpoint_dir += '_dct'
modelfile = get_best_file(checkpoint_dir)
print(checkpoint_dir)
