I found that no 'TEXT_EMB_PATH' and no 'OPENSET_TEST_TEXT_EMB_PATH' are not effective for transferring learning
SamXiaosheng opened this issue · comments
Sam commented
Hi, I am readlly interseted in your work, I found that no 'TEXT_EMB_PATH' and no 'OPENSET_TEST_TEXT_EMB_PATH' are not effective for transferring learning. so dose it not use text embeding?
Yiwu Zhong commented
Hi @SamXiaosheng, thanks for your interest. We did use text emb during transfer learning. I don't quite understand your question here. Could you elaborate on it?
Sam commented
Yiwu Zhong commented
@SamXiaosheng If you comment out the last two rows, it will be a closed-set training, such as a regular faster rcnn. Your model can only predict the base categories seen in training.