search

gaohan-12 / SPME

Structure-Preserving Motion Estimation for Learned Video Compression
13 stars 1 forks source link

All frames of the test sequence #2

Open xuezhongcailian opened 1 year ago

xuezhongcailian commented 1 year ago

Hello, the author, thank you for your work and code open source. Excuse me, in the test set, GOP = 12, if the test sequence is not an integer multiple of 12, I see in your code dataset.py, is it abandoned the frame behind?

gaohan-12 commented 1 year ago

Hello, the author, thank you for your work and code open source. Excuse me, in the test set, GOP = 12, if the test sequence is not an integer multiple of 12, I see in your code dataset.py, is it abandoned the frame behind?

Hi, the GoP size can be decided by yourself. The reason we set it to 12 here is to keep pace with prior works in leared video compression.

xuezhongcailian commented 1 year ago

Sorry, I want to say, the GOP settings, the video frame is not necessarily an integer multiple of the GOP, there will be a remainder, how to deal with it? thanks

gaohan-12 commented 1 year ago

Sorry, I want to say, the GOP settings, the video frame is not necessarily an integer multiple of the GOP, there will be a remainder, how to deal with it? thanks

Gop size is set to 10 and 12 for HEVC and UVG, respectively. And the number of frame in a sequence of the two datasets are actually integer multiples of 10 and 12. So there is no remainder.