novelvqa

novelvqa copied to clipboard

Something wrong when computing loss

3

 It goes wrong when computing loss in losses.py point_loss_ = self._point_loss(logits, pos.squeeze(1)) And I check the value of logits and pos  The shape of logits is (64,42), pos...

llipa

Hi, the reference set is unlabeled, so , dose it belong to training set or testing set? if it belongs to training set why it has the ability for novel...

hackerchenzhuo

hi，I would like to ask if it is ok to run this program with 4 1080Ti?

staring-s

Thanks for the interesting work! According to Appendix A in the paper, there are 10 types of skill labels. However, the json files in [https://github.com/SpencerWhitehead/novelvqa#concept-and-reference-set-preprocessing](https://github.com/SpencerWhitehead/novelvqa#concept-and-reference-set-preprocessing) only contain three of them...

RenShuhuai-Andy

hello, when will the code and your Novel-VQA Evaluation be released?

1

hello, when will the code and your Novel-VQA Evaluation be released?

156aasdfg

hello, when will the code and your Novel-VQA Evaluation be released?

hannajiang