search

tkbadamdorj / CHD

7 stars 0 forks source link

YouTube HighLights Dataset #2

Open WSnieming opened 1 year ago

WSnieming commented 1 year ago

Hi,

I'm doing related work, but some videos in the YouTube Highlights dataset have been lost. I see that you used the dataset in your paper. Can you share it with me?

Thank you, Ming Nie.

tkbadamdorj commented 1 year ago

Hi,

Code and datasets for a related paper (Joint visual and audio learning for video highlight detection) are available here https://drive.google.com/file/d/1zeJxeusP-GbyLw-DgU5Pe44qSnpbUldj/view and here https://drive.google.com/drive/u/1/folders/18ZjBdh8qa4rUvBH1WhEZ3EjDUfVO3hDz .

You'll find the YouTube dataset among the datasets.

Best, Taivanbat

On Tue, Dec 27, 2022 at 10:49 PM WSnieming @.***> wrote:

Hi,

I'm doing related work, but some videos in the YouTube Highlights dataset have been lost. I see that you used the dataset in your paper. Can you share it with me?

Thank you, Ming Nie.

— Reply to this email directly, view it on GitHub https://github.com/tkbadamdorj/CHD/issues/2, or unsubscribe https://github.com/notifications/unsubscribe-auth/AA4NTUSKZPGH4O6IIKM3HZDWPPPFZANCNFSM6AAAAAATLAAWFQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

WSnieming commented 1 year ago

Hi,

I used the data you shared to do some experiments and achieved good results. I also hope to use some additional video data for experiments. I want to use the ResNet model in your article to extract video features. Can you share the pretrained model with me?

Best, Ming

e-apostolidis commented 1 year ago

Thanks for releasing the data and code of your previous work (it will be great if you can share the data for this work as well to assist reproducibility of your results). Did you adopted the same evaluation protocol in both of your works (i.e. the "Joint visual and audio learning for video highlight detection" (ICCV 2021) and "Contrastive Learning for Unsupervised Highlight Detection" (CVPR 2022))? In the previous work you mention runs for 10 different random seeds, so I am wondering if this setting holds true for the latter work as well, as you report different numbers for your ICCV 2021 work in your CVPR 2022 paper.