A blazing-fast JPEG 2000

[Misha Engel]

https://htj2k.com/

The standard is royalty free and brings lightning fast image compression and decompression.

Use it with your existing images and workflows: all the features of JPEG 2000 plus lossless transcoding. It is designed to be used by all the current JPEG 2000 family of users such as D-Cinema, video post-production, film preservation, medical, satellite, etc.

Sound pretty good to me.

 

[Blair S. Paulsen]

Great link Misha. The HTJ2K variant of JPEG 2000 has a lot going for it and could be a legit option for the RED ecosystem. One of the caveats of the HTJ2K topology is lost support for low res extractions from discrete data. In other words, you could't just fetch the 1/4 (or 1/2, 1/16th, etc) res chunk and ignore the rest like you can with the nested structure of J2K. Until recently, losing that feature would have made working with R3Ds too heavy for a lot of projects.
These days it's a different equation. If, in fact, HTJ2K can achieve a 10x improvement in speed - leveraged with the latest silicon - we wouldn't need no stinking' extractions, proxy workflows, etc. I have no idea how many steps remain between what's in the white paper and actual wide implementation, but the potential is very interesting.

Cheers - #19

 

[Michael Lindsay]

Good link... thanks for the low res extraction info Blair...another example of one of the greatest strengths also being a weakness!

Just in case anyone from Red reads this...I haven’t decoded a r3d file any less than full debayer 16bit for as long as I can remember....

 

[Gunleik Groven]

Great stuff.

Thanks!

 

[Hrvoje Simic]

It says here encoding 2 fps and decoding 3 fps for 4K 4:4:4 on Core I7 6700.


https://github.com/aous72/OpenJPH

 

[Blair S. Paulsen]

It says here encoding 2 fps and decoding 3 fps for 4K 4:4:4 on Core I7 6700.
https://github.com/aous72/OpenJPH

Presumably they are still optimizing.

Cheers - #19

 

[David Rasberry]

What does this do for you that Cineform doesn’t?

 

[Blair S. Paulsen]

The orthodoxy has always been that J2K variants require modest encoding effort, but pretty serious decoding muscle. AFAIK, Cineform balances that out a bit which makes decoding faster, at the expense of needing more resources at encode. Since encode happens in camera, it can be the enemy of small/light/power sipping/etc.

Cheers - #19

 

[Misha Engel]

What does this do for you that Cineform doesn’t?

Cineform(RAW) is superior and that is the main reason commercial parties don't want to use it and stick to their own(developed) codecs, this might give them a way out and present us something better than they can offer us now (a bit like AV1 instead of HVEC, etc..).