When I use Python 3 training and do not use the instruction "- target", the following error occurs. What is the reason for kneeling

header0 label [1. 2.]
id2range 1
0 0 2
c2c_stat [0, 1]
0
rand_mirror 1
(3,)
Traceback (most recent call last):
File "train.py", line 1025, in
main()
File "train.py", line 1022, in main
train_net(args)
File "train.py", line 887, in train_net
data_set = verification.load_bin(path, image_size)
File "eval/verification.py", line 185, in load_bin
bins, issame_list = pickle.load(open(path, 'rb'))
UnicodeDecodeError: 'ascii' codec can't decode byte 0xff in position 0: ordinal not in range(128)

the commond is CUDA_VISIBLE_DEVICES='0' python -u train.py --network r100 --data-dir /home/xl/1mxnetinsight/insightface-master/datasets/17_faces_datasets --ckpt 2 --version-se 0 --per-batch-size 3 --prefix /media/xl/绿叶启动盘/save/model --verbose 10 --lr 0.00001

Start with train.py at line 887, where verification.load_bin is called, then inspect eval/verification.py at line 185 and the path supplied by --data-dir. Reproduce the command with Python 3 and examine the pickle input that triggers the UnicodeDecodeError. Done means training without the stated instruction no longer fails at pickle.load.

Written by the indexing model from the issue text.