The ‘msvd_inceptivev4.hdf5’ I extracted is different from the one you provided, I want to know what caused it?

hobincar / RecNet

A Pytorch implementation of "Reconstruction Network for Video Captioning", CVPR 2018

MIT License

54 stars 19 forks source link

The ‘msvd_inceptivev4.hdf5’ I extracted is different from the one you provided, I want to know what caused it? #6

Closed xiaoyongbest1234 closed 5 years ago

xiaoyongbest1234 commented 5 years ago

Hi, why did I extract msvd_inceptivev4.hdf5 with the size of 10.3GB using the pytorch-video-feature extractor, while you only provided 634.4MB? My MSVD dataset was downloaded fromhere. Is there an exception problem with the MSVD datasetI downloaded?Or did some of my actions cause 10.3GB? In addition, can you provide a link to download the MSVD/ MSR-VTT dataset?

I look forward to your reply and thank you very much.

hobincar commented 5 years ago

Hi,

I extract the InceptionV4 features for MSVD again by just cloning the repository to make sure I have the same code with you, and I got an hdf5 file of size 606MB. It is still different from what I already uploaded (I extracted the 636MB features a long time ago), but the 10.3GB features you got seems to be not normal.

I downloaded the MSVD dataset from here (which is the same as yours), and the MSR-VTT dataset from here.

To make sure anyone can extract features themselves from the repo, I'll upload the newly extracted features of size 606MB after I check its performance.

xiaoyongbest1234 commented 5 years ago

Thank you very much for your help. I will try again.

hobincar commented 5 years ago

@xiaoyongbest1234 I uploaded the MSVD and MSR-VTT Inception-v4 features in README.md. Please check :)