resize of the I frames P frames and B frames by a certain factor ?

[ocmtime]

How to (reduce the pixel count size) resize of the I frames P frames and B frames by a certain factor ?

If I frames are jpeg compressed and the blocksthemselves would be reduced to half resolution and the I frames to half resolution that would already be effective but how please ?

Also the other types of frames P and B containing motion vectors and plain difference between frames those would also need to be halved in resolution and there would be no necesity of recomputing the motion vectors because motion vectors are very robust to resolution

[videoh]

MPEG-2 does not use JPEG compression.

What exactly are you wanting to do? Reduce the size of an existing MPEG-2 file? If so, transcoding to HEVC would seem more rational.

[ocmtime]

I need to do that for a few terabytes of data and on my computer to hevc the speed is realtime at best. I need it to happen at 10x realtime speed.

[videoh]

Get a decent nVidia GPU and transcode with NVEnc. What percent size reduction are you looking for?

It's much easier to buy more storage. ;)

[manolito]

@ ocmtime
Your goal seems to be similar to what available CDT (Compressed Domain Transcoder) software already does (DVDShrink or DVD2one). The motion vectors are reused, but the original frames are recoded using a higher quantizer. Plus some other smart tweaks of course.