pytorch-vsumm-reinforce
pytorch-vsumm-reinforce copied to clipboard
Compare Ground truth and Generated Summary
Hi I have generated summary through tvsum and summe datasets. But I want to compare ground truth and generated summary frame by frame. Please give me any suggestions how I can start with it. If you need any more information, please reply here.
Hi, @n9839950. you can get a "gt_summary" in *.h5. There is a generated summary in result.json. so, you can compare its frame by frame.
thanks for the reply @SinDongHwan I am having one more problem. I have generated summaries on summe datasets. I have noticed that all my summary length are just 8 sec long. Technically it should be 15% of the original length. I am confused on this part.
Hello,I want to ask when I get "gt_summary"(tvsum video 10) is as follows, [1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. ....... 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 1. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0.] and "gtscore" is as following [0.3 0.3 0.3 0.3 0.25 0.25 0.25 0.25 0.1375 0.1375 0.1375 0.1375 0.125 0.125 0.125 0.125 0.1625 0.1625 0.1625 0.1625 ...... 0.2 0.2 0.2 0.2 0.1625 0.1625 0.1625 0.1625 0.125 0.125 0.125 0.125 0.2 0.2 0.2 0.2 0.2 0.2 ] How do you know which frame or segment "1" in "gtsummary" represents? How do you know which frame or segment is included in the final "user_summary"
@wangsen-sen hi, have you solve these problem, I am also confused that how the gtsummary get from 'user summary', or there exist another method to get gtsummary. As some papers claimed, gt_summary coule be directly obtained. lookiing forward your reply
@weiwuxian1998
@wangsen-sen
the size of array of gt_summary and gtscore should be same and then you can visualize it accordingly.
below screenshot shows red = 1 and blue = 0. And gtscore on yaxis.