Questions about the metrics in the paper
RobertLuo1 opened this issue · comments
Robert Luo commented
Eason commented
The same confusion above mentioned
Kevin commented
Hi @RobertLuo1 @EasonXiao-888
Thanks for asking!
The inconsistent result here is because the offical Moment_DETR do not report the R@0.3 and mIoU,
and I rerun their codebase on this dataset based on their recommended parameter settings (w/o HT100M PT).
I saved the best one based on the mIoU metrics (which is more comprehensive than R@0.5), thus the R1@0.5 may slightly lower even with best mIoU metrics.
For the second question, I evaluate in TaCOS test set. Please refer my metadata and features vid.
Robert Luo commented
I got it and thanks for your reply!!