mgong2 / da_infer Goto Github PK

In the digits experiments, 20,000 paired images are sampled from the training set for each dataset. But the syn_digits dataset only contain 10000 samples in SYN_DIGITS_train.pt. Is there any solution？

Error happened in line 80 of code_digits/train.py:
ValueError: Cannot take a larger sample than population when 'replace=False'

In code_digits/twin_ac_launch_mnist_mnist.sh, the setting includes:
--source_dataset svhn,mnist_m,mnist,syn_digits --target_dataset svhn.

It seems that target domain svhn's labels are used while training (see line 283 and 354 in train.py). Is there any mistakes?
If there is any implementations to exclude labels of svhn while training, please tell me. Thank you.