def merge():
splits_id_path = '../data/splits'
vcoco_train = np.loadtxt(os.path.join(splits_id_path, 'vcoco_train.ids'))
print(len(vcoco_train))
train_list = [str(int(id)) for id in vcoco_train]
vcoco_val = np.loadtxt(os.path.join(splits_id_path, 'vcoco_val.ids'))
print(len(vcoco_val))
val_list = [str(int(id)) for id in vcoco_val]
vcoco_test = np.loadtxt(os.path.join(splits_id_path, 'vcoco_test.ids'))
print(len(vcoco_test))
test_list = [str(int(id)) for id in vcoco_test]
vcoco_trainval = np.loadtxt(
os.path.join(splits_id_path, 'vcoco_trainval.ids'))
print(len(vcoco_trainval))
trainval_list = [str(int(id)) for id in vcoco_trainval]
d = {}
d["train"] = train_list
d["val"] = val_list
d["test"] = test_list
d["trainval"] = trainval_list
stats = {}
stats["test"] = len(test_list)
stats["train"] = len(train_list)
stats["trainval"] = len(trainval_list)
stats["val"] = len(val_list)
io.dump_json_object(d, 'data_process/split_ids_all.json')
io.dump_json_object(stats, 'data_process/split_ids_stats_all.json')
def save_hois_to_json(data_dict):
data_dict = dict(
sorted(data_dict.items(), key=lambda d: d[0], reverse=False))
id = 1
data_list = []
rare_list = []
non_rare_list = []
for action_ob_key in data_dict.keys():
name = action_ob_key.split('/')
assert len(name) == 2
action = name[0]
ob = name[1]
set_num_dict = data_dict[action_ob_key]
temp = {
'id': str(id).zfill(3),
'verb': action,
'object': ob,
'train_num': set_num_dict.get('train', 0),
'val_num': set_num_dict.get('val', 0),
'test_num': set_num_dict.get('test', 0),
'rare': True if set_num_dict.get('train', 0) < 10 else False
}
if temp['rare']:
rare_list.append(id)
else:
non_rare_list.append(id)
id = id + 1
data_list.append(temp)
print(data_list)
print(type(data_list))
print(type(data_list[0]))
file = os.path.join(os.getcwd(), 'hois', 'hoi_list.json')
io.dump_json_object(data_list, file)
def check():
anno_file = os.path.join("data_process", 'anno_list_gt.json')
anno_list = io.load_json_object(anno_file)
global_ids = [anno['global_id'] for anno in anno_list]
print(len(global_ids))
ids = io.load_json_object(
os.path.join("data_process", 'split_ids_all.json'))
train_lost = []
val_lost = []
test_lost = []
train_list = []
val_list = []
test_list = []
trainval_list = []
for id in ids['train']:
if not id in global_ids:
print("train", id)
train_lost.append(id)
else:
train_list.append(id)
trainval_list.append(id)
for id in ids['val']:
if not id in global_ids:
print("val", id)
val_lost.append(id)
else:
val_list.append(id)
trainval_list.append(id)
for id in ids['test']:
if not id in global_ids:
print("test", id)
test_lost.append(id)
else:
test_list.append(id)
d = {}
d['train'] = train_lost
d["val"] = val_lost
d['test'] = test_lost
io.dump_json_object(d, 'data_process/id_lost.json')
d = {}
d['train'] = len(train_lost)
d["val"] = len(val_lost)
d['test'] = len(test_lost)
io.dump_json_object(d, 'data_process/id_lost_stats.json')
d = {}
d['train'] = train_list
d["val"] = val_list
d['test'] = test_list
d['trainval_list'] = trainval_list
io.dump_json_object(d, 'data_process/split_ids.json')
d = {}
d['train'] = len(train_list)
d["val"] = len(val_list)
d['test'] = len(test_list)
d['trainval_list'] = len(trainval_list)
io.dump_json_object(d, 'data_process/split_ids_stats.json')
category_ids_list2 = merge_two_list(trainval_list[4], test_list[4])
name_list2 = get_names(coco, category_ids_list2)
# print(name_list2)
print("1.Action:", trainval_list[6])
print('2.Roles:', "1" if len(trainval_list[5]) == 2 else "2")
print("3.numbers of example:", trainval_list[0] + test_list[0])
result = {}
for i in range(1, len(trainval_list[5])):
print("4.", trainval_list[5][i])
print("5.", trainval_list[1] + test_list[1] if i == 1 else trainval_list[2] + test_list[2])
print("6.Objects in role:", trainval_list[5][i], name_list1 if i == 1 else name_list2)
result[trainval_list[5][i]] = name_list1 if i == 1 else name_list2
dict[trainval_list[6]] = result
return dict
if __name__ == '__main__':
coco = COCO(os.path.join('coco/annotations/', 'instances_trainval2014.json'))
v_coco = vu.load_coco()
dict = {}
# get_table1_one_row(action='sit')
for i in range(26):
result = get_table1_one_row(v_coco, coco, action_index=i, dict=dict)
assert len(dict) == 26
io.dump_json_object(dict, "table_list.json")
def generate_json_data_for_vcoco_total_with_gt():
hoi_list = json.load(
open(os.path.join(os.getcwd(), 'data_process', 'hoi_list.json'), 'r'))
coco = COCO(
os.path.join('..', 'coco/annotations/', 'instances_trainval2014.json'))
anno_list = []
verb_set = set()
object_set = set()
for dataset_name in ["vcoco_train", "vcoco_val", "vcoco_test"]:
vcoco_data = vu.load_vcoco(dataset_name)
# train_data add bbox and role_bbox
for x in vcoco_data:
x = vu.attach_gt_boxes(x, coco)
assert len(vcoco_data) == 26
for data in vcoco_data:
# print(data.keys())
verb_set.add(data['action_name'])
image_ids = data['image_id']
for i in range(len(image_ids)):
assert data['role_object_id'][i][0] == data['ann_id'][i]
if data['label'][i] == 0:
continue
path = 'train' if 'train' in data['file_name'][i] else 'val'
global_id = "COCO_" + path + "2014_" + str(
image_ids[i][0]).zfill(12)
image_path_postfix = os.path.join(path + '2014',
global_id + '.jpg')
global_id = str(image_ids[i][0])
human_bboxes = data['bbox'][i]
assert len(human_bboxes) == 4
# get image size
image_size = data['image_size'][i]
assert len(image_size) == 2
hois = []
assert len(data['role_name']) <= 3
for j in range(1, len(data['role_name'])):
object_anno_id = data['role_object_id'][i][j]
if object_anno_id == 0:
continue
d = {}
object_anno = coco.loadAnns(int(object_anno_id))[0]
object_bboxes = data['role_bbox'][i][4 * j:4 * (j + 1)]
cat_id = object_anno['category_id']
object_name = coco.loadCats(int(cat_id))[0]['name']
object_set.add(object_name)
id = -1
for k in hoi_list:
if k['action'] == data['action_name'] and k[
'object'] == object_name:
id = k['id']
break
assert id > 0
d['object'] = object_name
d['human_bboxes'] = human_bboxes
d['id'] = id
d['invis'] = 0
d['action'] = data['action_name']
d['object_bboxes'] = object_bboxes
hois.append(d)
if hois == []:
# print(i, data["action_name"], data['label'][i], data['role_object_id'][i])
continue
d_per_image = {}
d_per_image['global_id'] = global_id
d_per_image['hois'] = hois
d_per_image['image_path_postfix'] = image_path_postfix
d_per_image['image_size'] = image_size
anno_list.append(d_per_image)
print(dataset_name, len(anno_list))
anno_dict = {}
print(len(anno_list))
for item in anno_list:
assert item['hois'] is not None
if item['global_id'] not in anno_dict:
anno_dict[item['global_id']] = item
else:
anno_dict[item['global_id']]['hois'].append(item['hois'][0])
print(len(anno_dict))
anno_list = []
for _, value in anno_dict.items():
anno_list.append(value)
io.dump_json_object(anno_list, 'data_process/' + 'anno_list_gt.json')
ob_index = 1
object_set = sorted(object_set)
object_list = []
for ob in object_set:
object_list.append({"id": str(ob_index).zfill(3), "name": ob})
ob_index += 1
io.dump_json_object(object_list, 'data_process/' + 'object_list_gt.json')
vb_index = 1
verb_set = sorted(verb_set)
verb_list = []
for verb in verb_set:
verb_list.append({"id": str(vb_index).zfill(3), "name": verb})
vb_index += 1
io.dump_json_object(verb_list, 'data_process/' + 'verb_list_gt.json')