search

boheumd / A2Summ

The official implementation of 'Align and Attend: Multimodal Summarization with Dual Contrastive Losses' (CVPR 2023)
https://boheumd.github.io/A2Summ/
71 stars 10 forks source link

How to get the labels for the multimodel summary? #13

Open eMoLii opened 11 months ago

eMoLii commented 11 months ago

Thx for your great work！But I am confused by the labels in the multimodel datasets(CNN and Daily Mail).How did you get the labels for the summary?I have read the aritle for many times, but I can't find anything about it. Thx for your reply!

boheumd commented 11 months ago

Hi, thank you for your interest in our work. You can refer to the original data through this GitHub repo https://github.com/xiyan524/MM-AVS. For the text summary, CNN and DailyMail already contain the ground-truth annotation in highlight.txt. For the video summary, we first extract the frames for videos and find the most similar frames to each ground-truth image (*.png) in each data sample.

eMoLii commented 11 months ago

Thank you for your reply! I already know the video summary lables through your answer. But I am still confusing about the text summary lables. CNN and DailyMail do have highlight.txt. But it is the abstract summary written by human, not the extract summary label we need for training the model.

boheumd commented 11 months ago

Sorry for the mistake. I checked for the text summary, the ground-truth labels are already provided in the original data. I list the label file link for the CNN dataset and label file link for the Daily Mail dataset here. It's in the original OneDrive repo.

eMoLii commented 11 months ago

Thx again. I already find them. perblem solved!