Synopsis

Computational cinematography software for content creators and film makers.

By Special Circumstances. @_vade & @rsomani95

[Anton Marini @_vade]

Have a Panel UI for Synopsis analysis JSON parsing in After Effects now. This is a nice (albeit slow) environment to debug but also explore inference results.

Really proud of the work @rsomani95 has been doing on improving our model. Note how the composition tracks focus A+.

[Anton Marini @_vade]

I really cant wait to run @_synopsis_ on M1 Max and see how ridiculous it is.

[Creative Visual Generation Editing Understanding @cveu_workshop]

Here is a great Snapshot of our Industry Speakers taken by @raoanyi!
Joon-Young, @agermanidis, @_vade, @rsomani95, Fernando Amat, Xintao Wang

[Creative Visual Generation Editing Understanding @cveu_workshop]

Special thanks to our Industry Speakers!
Joon-Young, @AdobeResearch
@agermanidis, @runwayml
@_vade @rsomani95, @_synopsis_
Fernando Amat, @NetflixResearch
Xintao Wang, Tencent!
Great insights coming from real-world applications!
#iccv2021 @ICCV_2021

[Anton Marini @_vade]

This wouldn’t have been possible without @_synopsis_ - which has come a long way and is putting in some serious work these days.

[Anton Marini @_vade]

We put up an article detailing the insane amount of work done for @hbomax by @HushStudios and Special Circumstances (@rsomani95 and my company which doesn’t have yet have a twitter)

If you’re curious about #ai / #ml and #film / #cinematography check It:

https://t.co/dVQJutzh5q

[Anton Marini @_vade]

Really proud of the work @rsomani95 and I did for cinematic @_synopsis_ analysis of @hbomax ’s content library on behalf for our friends at @HushStudios. The entire team worked really hard on this and it came out lovely.

New & improved models powered:

https://t.co/9ga4clDMy7

[Anton Marini @_vade]

@_synopsis_ @rsomani95 All from the same external SSD

Hackintosh w/ i9 10k 128GB Radeon VII: ~ 1:30s

MacBook Pro w eGPU 580 : ~2:15s

M1 Mac Mac mini: ~15s

No. That is not a typo. Maybe theres a new case of unified memory bugs my threading code fails on but this output seems ok?

[Anton Marini @_vade]

So, I just got @_synopsis_ running natively on M1 Apple Silicon.
We have the 16GB Mac mini.

A 4 minute and 30 second 720p h.264 movie analyzed with a CoreML model, as well as some custom metal performance shaders:

Cc @rsomani95

[Anton Marini @_vade]

Sneak peek: we’ve been quietly working on improving @_synopsis_ - @rsomani95 has been making huge progress on our new location data set / model. These images are sorted against only shots at the same inferred *location* :

[Synopsis @_synopsis_]

@jorge2k @_vade Link is pinned on our slack - apologies we’ve been re-working some things internally and haven’t had a chance to update links / releases!

https://t.co/SV0Wu7h4Xf

[Synopsis @_synopsis_]

@walidvb We have some workflows for editors and scripts we are working on :)

[Synopsis @_synopsis_]

@editblog Interesting - what was the file that caused the crash?

[Synopsis @_synopsis_]

@rsomani95 @_vade Ocean: Moonlight

[Synopsis @_synopsis_]

Testing drastically improved classifiers from Fast AI into CoreML - work done by @rsomani95 / @_vade

Moonlight - car shots:

[Synopsis @_synopsis_]

@rsomani95 @_vade Master shots, Close Up shots, from Moonlight:

[Synopsis @_synopsis_]

@2bitpunk Haha awesome. And yes, theres never enough storage!

[Synopsis @_synopsis_]

@2bitpunk If you are up for it, join our slack or post a bug / feature request at https://t.co/oVvksurSVv

Def let us know if the crashes are repeatable!

[Synopsis @_synopsis_]

@2bitpunk Ah interesting! Thats a great point. Right now, CoreML and most ML tooling actually doesnt support alpha channel (tensors are ranked and each dimension has an assumed role sadly - the 4th channel is for ‘batches’, not alpha! Argh!)

However a nice UI would be great!