Some results issue
viyjy opened this issue · 4 comments
Hi Junnan, I have the following questions about the result. Hope that you can help to clarify them, thanks.
-
VQA:
I get theresult
folder after fine-tuning on VQA dataset. Which json file should I use to get test-dev and test-std?
-
SNLI-VE:
This is the log file after fine-tuning on SNLI-VE dataset. You didn't update the best-epoch, so it is always 0. Should I pick the row which has the best val accuracy as the final result?
-
Grouding
This is the log file after fine-tuning on Ref-COCO. Should I pick the row which has the best val_d as the final result?
-
NLVR2
This is the log file after fine-tuning on NLVR2, but I did't finddev
andtest-P
as shown in your paper, any idea?
Hi, thanks for your interest. Here are my answers.
- VQA: vqa_result_epoch7.json is the final result which collects results from all ranks.
- SNLI-VE: thanks for spotting my mistake. Yes you should use the row with best val_acc.
- Grounding: if I remembered correctly, in the paper I just reported the last epoch's result.
- NLVR: dev is val, test-P is test.
Thanks very much!
Hello! Can NLVR2 datasets be shared?
You can find it here: https://github.com/lil-lab/nlvr/tree/master/nlvr2