def evm_train_process(classes_to_process, feature_dic, args_evm, gpu, Q,
done_event):
with torch.no_grad():
evm_iterator_i = EVM_Training(classes_to_process, feature_dic,
args_evm, gpu)
evm_model = OrderedDict()
n = 0
for evm in enumerate(evm_iterator_i):
# evm_model[evm[1][1][0]] = evm[1][1][1]
Q.put((evm[1][1][0], evm[1][1][1]))
gc.collect()
torch.cuda.empty_cache()
torch.cuda.ipc_collect()
done_event.wait()
del evm
return
for j, im_name in enumerate(image_names):
label = feedback.get_label(im_name)
if label > number_of_known_classes:
class_to_process.add(label)
feature_j = FV[j : (j + 1), :].clone().double()
if label in features_dict_train:
features_dict_train[label] = torch.cat((features_dict_train[label], feature_j), 0)
else:
nu = nu + 1
features_dict_train[label] = feature_j
if len(class_to_process) > 0:
class_to_process = sorted(list(class_to_process))
list_of_tuples = [0.0] * len(class_to_process)
evm_iterator_i = EVM_Training(class_to_process, features_dict_train, args_evm_unknown, 0)
evm_counter = 0
for evm in enumerate(evm_iterator_i):
# label = evm[1][1][0]
# mini_evm = evm[1][1][1]
list_of_tuples[evm_counter] = (evm[1][1][0], evm[1][1][1])
evm_counter = evm_counter + 1
for j, class_number in enumerate(class_to_process):
# evm is an OrderedDict
assert list_of_tuples[j][0] == class_number
evm_model[class_number] = list_of_tuples[j][1]
del evm_iterator_i, evm
gc.collect()
torch.cuda.empty_cache()
to_be_delete = to_be_delete + index
nu = nu + 1
class_number = int(nu + number_of_discovered_classes + number_of_known_classes)
features_dict_incremental[class_number] = (
torch.from_numpy(np.array([FVs_residual[jjj] for jjj in index]))
).double() # .cuda()
class_to_process.append(class_number)
if ((len(class_to_process) > 0) and (number_of_discovered_classes > 0)) or (
len(class_to_process) > 2
):
list_of_tuples = [0.0] * len(class_to_process)
evm_iterator_incremental = EVM_Training(
class_to_process, features_dict_incremental, args_evm_incremental, 0
)
evm_counter = 0
for evm in enumerate(evm_iterator_incremental):
# label = evm[1][1][0]
# mini_evm = evm[1][1][1]
list_of_tuples[evm_counter] = (evm[1][1][0], evm[1][1][1])
evm_counter = evm_counter + 1
for j, class_number in enumerate(class_to_process):
# evm is an OrderedDict
assert list_of_tuples[j][0] == class_number
evm_model_incremental[class_number] = list_of_tuples[j][1]
del evm_iterator_incremental, evm
gc.collect()