gulvarol / ltc

Long-term Temporal Convolutions for Action Recognition, TPAMI 2018
http://www.di.ens.fr/willow/research/ltc
MIT License
85 stars 26 forks source link

How long it takes to compute optical flow for UCF101 dataset? #7

Open LongLong-Jing opened 5 years ago

LongLong-Jing commented 5 years ago

@gulvarol

Hi, I want to compute optical flow for UCF101 datasets. Do you remember how long it takes? Thanks for your help.

gulvarol commented 5 years ago

It takes ~0.2 seconds to extract 1 frame with GPU implementation of Brox optical flow. As far as I remember, I parallelized on 2 GPUs and extracted in ~1 day the UCF101 dataset. Now, you can use more recent flow algorithms.

LongLong-Jing commented 5 years ago

Thanks!

On Wed, Mar 6, 2019 at 7:44 AM Gul Varol notifications@github.com wrote:

It takes ~0.2 seconds to extract 1 frame with GPU implementation of Brox optical flow. As far as I remember, I parallelized on 2 GPUs and extracted in ~1 day the UCF101 dataset. Now, you can use more recent flow algorithms.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/gulvarol/ltc/issues/7#issuecomment-470093729, or mute the thread https://github.com/notifications/unsubscribe-auth/AUWi5amB0zmShrYJwe72y2Ekk_PDyS98ks5vT7g9gaJpZM4bf6UF .

-- Best, Longlong

Longlong Jing, Ph.D. Candidate. Media Lab @ CUNY City College. 160 Convent Avenue, New York, NY 10031 https://maps.google.com/?q=160+Convent+Avenue,+New+York,+NY+10031&entry=gmail&source=g .

https://longlong-jing.github.io/ https://longlong-jing.github.io/longlongjing.cs@gmail.com longlongjing.cs@gmail.com