def vidvrd_generator(self):
feature_type = 'train'
# for each_ins in vid_data.gen_vrd_instance(feature_type):
# yield {
# "image/encoded": np.array2string(each_ins.get_my_feature(feature_type).ravel()),
# "image/format": ["jpeg"],
# "image/class/label": [vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]],
# "image/height": [each_ins.height],
# "image/width": [each_ins.width]
# }
ins_list = vid_data.gen_vrd_instance(feature_type)
for i in range(3):
yield {
"image/encoded":
np.array2string(
ins_list[i].get_my_feature(feature_type).ravel()),
"image/format": ["jpeg"],
"image/class/label": [
vid_relation.load_relation('first')[
ins_list[i].predicate.split('_')[0]]
],
"image/height": [ins_list[i].height],
"image/width": [ins_list[i].width]
}
def feature_encoders(self, data_dir):
del data_dir
feature_type = 'train'
for each_ins in vid_data.gen_vrd_instance(feature_type):
yield {
"inputs": each_ins.get_my_feature(feature_type),
"targets": vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]
}
def vidvrd_generator(self):
feature_type = 'train'
for each_ins in vid_data.gen_vrd_instance(feature_type):
yield {
"image/encoded": np.array2string(each_ins.get_my_feature(feature_type).ravel()),
"image/format": ["png"],
"image/class/label": [vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]],
"image/height": [each_ins.height],
"image/width": [each_ins.width]
}
def vidvrd_generator(self):
feature_type = 'train'
# for each_ins in vid_data.gen_vrd_instance(feature_type):
# features = {"image/encoded": np.array2string(each_ins.get_my_feature(feature_type).ravel()),
# "image/format": ["png"],
# "image/height": [each_ins.height],
# "image/width": [each_ins.width]}
# yield features
ins_list = vid_data.gen_vrd_instance(feature_type)
for i in range(3):
features = {"image/encoded": np.array2string(ins_list[i].get_my_feature(feature_type).ravel()),
"image/format": ["jpg"],
"image/height": [ins_list[i].height],
"image/width": [ins_list[i].width]}
yield features
def get_vvrd_truth_list(data_type, relation_type):
pos = 0 if relation_type == 'first' else 1
file_path = '../data/vvrd_truth_' + data_type + '_' + relation_type + '_list.json'
vrd_json = {}
with open(file_path, 'w+') as f:
for each_vrd in get_data.gen_vrd_instance(data_type):
if pos == 1:
if len(each_vrd.predicate.split('_')) > 1:
rel = load_relation(relation_type)[
each_vrd.predicate.split('_')[pos]]
else:
rel = -1
else:
rel = load_relation(relation_type)[each_vrd.predicate.split(
'_')[pos]]
vrd_json[each_vrd.video_id + '_' + str(each_vrd.begin_fid) + '_' +
str(each_vrd.end_fid)] = rel
# each_vrd_json = {each_vrd.video_id + '_'
# + str(each_vrd.begin_fid) + '_'
# + str(each_vrd.end_fid): rel}
f.write(json.dumps(vrd_json))
return 'Successfully get VVRD Truth List: ' + file_path
"targets": vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]
}
def vidvrd_generator(self):
feature_type = 'train'
for each_ins in vid_data.gen_vrd_instance(feature_type):
yield {
"image/encoded": np.array2string(each_ins.get_my_feature(feature_type).ravel()),
"image/format": ["png"],
"image/class/label": [vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]]],
"image/height": [each_ins.height],
"image/width": [each_ins.width]
}
# ins_list = vid_data.gen_vrd_instance(feature_type)
# for i in range(3):
# yield {
# "image/encoded": np.array2string(ins_list[i].get_my_feature(feature_type).ravel()),
# "image/format": ["png"],
# "image/class/label": [vid_relation.load_relation('first')[ins_list[i].predicate.split('_')[0]]],
# "image/height": [ins_list[i].height],
# "image/width": [ins_list[i].width]
# }
if __name__ == '__main__':
feature_type = 'train'
each_ins = vid_data.gen_vrd_instance(feature_type)[0]
print(each_ins.get_my_feature(feature_type))
print(vid_relation.load_relation('first')[each_ins.predicate.split('_')[0]])