From: "Nick Fox-Gieg" Received: from mail-io1-f50.google.com ([209.85.166.50] verified) by media-motion.tv (CommuniGate Pro SMTP 6.1.0) with ESMTPS id 7248392 for AE-List@media-motion.tv; Thu, 21 Feb 2019 19:48:57 +0100 Received: by mail-io1-f50.google.com with SMTP id p196so3101635iod.9 for ; Thu, 21 Feb 2019 10:59:37 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to; bh=nC6CqRXkxw9gSyouGDKUq1EEd85lH2L05xAcFzGQV0U=; b=odUBQKgMonLhPMjYMpmOddqP2O8gbFJ+bQHOeFr5fZweeML66No4LooOTuf4u6/hif 3uSlMBmn4a3AkY5xxovPs+nLjkk1AyY8HtzLvxyhqPYv9nGu6HhCrxf31PTdajbYZeRS ycinkPyQwXtAWxJoqQrpoN7sLNWV7CJuwnao3XRN+RoDXXrkqr6gg/2Kywt6pPirULW8 H6pAebvyTodgzkWqgRrBCiA9ARdsNttptWX0O36+rOQ+6A9ZF6MGvHxYr1Zg29pdtyuF CXQYUyPCHcRu+qLX0fsHSBVfjj6sOhnmNEuZ7vce0CrFeMJdwgxbuMPi6ymapqU/Cb5K lcMA== X-Gm-Message-State: AHQUAuYqXUu+7+Q4YRwBSbpLMDCuTrHQ3mnphxtE7xmxP99qaNz/zVZR 3u3u1k2txU5TmPrjyof6nqp4nuw15hbP5stsYti7Rw== X-Google-Smtp-Source: AHgI3IZzrDeEI8aSdNeGNA7zoBogidzRsTvsmW9G1C11cOrK+XaS3bGXB+mn2sg7v8lLwXJkEq9LIrllrrp2vq/jbMo= X-Received: by 2002:a6b:6b19:: with SMTP id g25mr16727ioc.152.1550775575747; Thu, 21 Feb 2019 10:59:35 -0800 (PST) MIME-Version: 1.0 References: In-Reply-To: Date: Thu, 21 Feb 2019 13:58:59 -0500 Message-ID: Subject: Re: [AE] Visually lossless codec thoughts To: After Effects Mail List Content-Type: multipart/alternative; boundary="00000000000051a03c05826c13eb" --00000000000051a03c05826c13eb Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Fwiw I find a lot of banding issues in ffmpeg can be solved with an initial ImageMagick pass to do the desired bit depth conversion before the final encode. Means a whole other set of arcane command-line switches to wrangle, but a powerful combo. Nick On Thu, Feb 21, 2019 at 1:48 PM Stephen van Vuuren wrote: > Chris, > > > > FFMPEG won=E2=80=99t resize the resize the various size PNGs for direct c= onversion > using it natively. > > > As for artifacts, PhotoJPEG in AE 2014 (the only version that will let me > render to it is 2014, missing it all other versions), crushed the blacks, > shift the colors to red which likely is causing the encoder to alter the > dithering in 8-bit i.e. rounding errors on a RGB value. > > > > Bottom line is very different image look where as all other codecs tried > are 100% visually identical. > > > > *From:* After Effects Mail List > *Sent:* Thursday, February 21, 2019 12:19 AM > *To:* After Effects Mail List > *Subject:* Re: [AE] Visually lossless codec thoughts > > > > Good catch--to clarify, the command switch for ffmpeg is mjpeg, but the > codec being used is PhotoJPEG. I think you have to specify 444 vs. 422 > encoding, instead of it kicking in automatically via the quality setting: > > https://trac.ffmpeg.org/wiki/Encode/VFX > > > > Nick > > > > On Wed, Feb 20, 2019 at 11:35 PM Chris Zwar > wrote: > > I'm not that familiar with ffmpeg, but with the older (now obsolete) Appl= e > Quicktime codecs, the MJPEG options were different to PhotoJPEG. The App= le > PhotoJPEG codec - even when used to make quicktime files - was a standard > JPEG codec. The MJPEGA & MJPEGB codecs date back to the early 1990s, whe= n > the only way to play back video on a desktop computer was with a custom > video card. The MJPEG codes were written for those custom chips. I can't > remember the exact details but it was something like MJPEGA was for C-Cub= e > chips and MJPEGB was for a different brand, or vice versa. The MJPEG > algorithms were different to Photo JPEG and not capable of the same level > of quality. I can't recall the details but I think the MJPEG codecs were > always 4:2:2 (ie colour information was half the bandwidth of the > luminance) while with the PhotoJPEG codec @ 100% the colour sampling was > 4:4:4. It's all a historical footnote now, anyway. > > > > FWIW HAP has it's purposes - a lot of the large scale projection projects > I work on deliver HAP files - but the quality is pretty poor and it's > nowhere near visually lossless. > > > > I'm not sure that something exists that meets all of Stephens > requirements. Without having done personal research, if you're not happy > with any of the ProRes flavours (have you tried LT?) then I don't know if > there's a viable alternative. > > > > > > > > On Thu, 21 Feb 2019 at 15:18, Nick Fox-Gieg > wrote: > > I've had good luck with ffmpeg's PhotoJPEG (aka mjpeg) in the past. For > 8bpc also look at Hap--large file sizes, uses GPU texture compression, > designed for very fast playback of > 4K. > > > > Nick > > > > On Wed, Feb 20, 2019, 11:06 PM Chris Zwar > Stephen what artefacts are you getting @ 100% quality? I've always > thought - after being directly told by an Apple engineer - that JPEG at > 100% was practically lossless, and any differences were due to rounding > errors and not compression artefacts. > > > > On Thu, 21 Feb 2019 at 13:33, Stephen van Vuuren > wrote: > > Teddy, > > > > PhotoJPEG was a fail =E2=80=93 color & gamma shift, artifacts at 100% qua= lity. > > > > Stephen > > > > *From:* After Effects Mail List > *Sent:* Wednesday, February 20, 2019 10:59 AM > *To:* After Effects Mail List > *Subject:* Re: [AE] Visually lossless codec thoughts > > > > Did you try photo-jpeg in an mov? Definitely worth a test. > > > > On Wed, Feb 20, 2019 at 10:48 AM Stephen van Vuuren < > AE-List@media-motion.tv> wrote: > > X264 lossless and x265 are both fails. X264 at 10-bit, AE says is 8-bit. > X265 take 10x time to render out in Media Encoder and takes an unknown ti= me > (at least 100x) to build a frame. I gave up after 11 hours =E2=80=93 ever= y other > codec was 14 to 22 minutes a frame. > > > > Adobe and x265 simply don=E2=80=99t play in my experience. > > > > *From:* After Effects Mail List > *Sent:* Wednesday, February 20, 2019 5:57 AM > *To:* After Effects Mail List > *Subject:* Re: [AE] Visually lossless codec thoughts > > > > x264 has lossless compression methods and can also be 10bit. > > AfterCodec has support for 10bit and YUV 4:4:4, not the lossless mode as > far as i can see, but a high bitrate x264 should be better than the old > simpler DCT based Prores or DNxHR at comparable bitrates > > > > Now the question is if AE will import them correctly again. > > > > https://trac.ffmpeg.org/wiki/Encode/H.264 > > > > Robert > > > > > > Den ons. 20. feb. 2019 kl. 09.08 skrev Evan Fotis Georgoulakis < > AE-List@media-motion.tv>: > > Personally I have the opposite experience, with Prores producing about > 20% larger files than same flavor Cineform. I tried CFYUV10bit. > > On 20-Feb-19 01:07, Teddy Gage wrote: > > Chris AFAIK cineform files are even bigger than Prores at the same > > flavor. > > > +---End of message---+ > To unsubscribe send any message to > > > > > -- > > _____________________________ > > Teddy Gage > > President > > Shotgun Inc. > > 360 VR | MOGRAPH | VFX > > SHOTGUNPOST.COM > > BKLYN NY 11215 > > --00000000000051a03c05826c13eb Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable
Fwiw I find a lot of banding issues in ffmpeg can be = solved with an initial ImageMagick pass to do the desired bit depth convers= ion before the final encode. Means a whole other set of arcane command-line= switches to wrangle, but a powerful combo.

Nick

On Thu, Feb 21, 2019 at 1:48 PM Stephen van Vuuren <<= a href=3D"mailto:AE-List@media-motion.tv">AE-List@media-motion.tv> w= rote:

Chris,

=C2=A0

FFMPEG won=E2=80=99t resize the resize the various s= ize PNGs for direct conversion using it natively.


As for artifacts, PhotoJPEG in AE 2014 (the only version that will let me r= ender to it is 2014, missing it all other versions), crushed the blacks, sh= ift the colors to red which likely is causing the encoder to alter the dith= ering in 8-bit i.e. rounding errors on a RGB value.

=C2=A0

Bottom line is very different image look where as al= l other codecs tried are 100% visually identical.

=C2=A0

From: After Effects Mail List <AE-List@media-motion.tv>
Sent: Thursday, February 21, 2019 12:19 AM
To: After Effects Mail List <AE-List@media-motion.tv>
Subject: Re: [AE] Visually lossless codec thoughts

=C2=A0

Good catch--to clarify, the command switch for ffmpe= g is mjpeg, but the codec being used is PhotoJPEG. I think you have to spec= ify 444 vs. 422 encoding, instead of it kicking in automatically via the qu= ality setting:

=C2=A0

Nick

=C2=A0

On Wed, Feb 20, 2019 at 11:35 PM Chris Zwar <AE-List@media-moti= on.tv> wrote:

I'm not that familiar with ffmpeg, but with the = older (now obsolete) Apple Quicktime codecs, the MJPEG options were differe= nt to PhotoJPEG.=C2=A0 The Apple PhotoJPEG codec - even when used to make q= uicktime files - was a standard JPEG codec.=C2=A0 The MJPEGA & MJPEGB codecs date back to the early 1990s, when the only= way to play back video on a desktop computer was with a custom video card.= =C2=A0 The MJPEG codes were written for those custom chips. I can't rem= ember the exact details but it was something like MJPEGA was for C-Cube chips and MJPEGB was for a different brand, or = vice versa.=C2=A0 The MJPEG algorithms were different to Photo JPEG and not= capable of the same level of quality.=C2=A0 I can't recall the details= but I think the MJPEG codecs were always 4:2:2 (ie colour information was half the bandwidth of the luminance) while with= the PhotoJPEG codec=C2=A0@ 100% the colour sampling was 4:4:4.=C2=A0 It= 9;s all a historical footnote now, anyway.

=C2=A0

FWIW HAP has it's purposes - a lot of the large = scale projection projects I work on deliver HAP files - but the quality is = pretty poor and it's nowhere near visually lossless.

=C2=A0

I'm not sure that something exists that meets al= l of Stephens requirements.=C2=A0 Without having done personal research, if= you're not happy with any of the ProRes flavours (have you tried LT?) = then I don't know if there's a viable alternative.

=C2=A0

=C2=A0

=C2=A0

On Thu, 21 Feb 2019 at 15:18, Nick Fox-Gieg <AE-List@media-moti= on.tv> wrote:

I've had good luck with ffmpeg's PhotoJPEG (= aka mjpeg) in the past. For 8bpc also look at Hap--large file sizes, uses G= PU texture compression, designed for very fast playback of > 4K.<= u>

=C2=A0

Nick=C2=A0

=C2=A0

On Wed, Feb 20, 2019, 11:06 PM Chris Zwar <AE-List@media-motion= .tv wrote:

Stephen what artefacts are you getting=C2=A0@ 100% q= uality?=C2=A0 I've always thought - after being directly told by an App= le engineer - that JPEG at 100% was practically lossless, and any differenc= es were due to rounding errors and not compression artefacts.

=C2=A0

On Thu, 21 Feb 2019 at 13:33, Stephen van Vuuren <= ;AE-List@media= -motion.tv> wrote:

Teddy,

=C2=A0

PhotoJPEG was a fail =E2=80=93 color & gamma shi= ft, artifacts at 100% quality.

=C2=A0

Stephen

=C2=A0

From: After Effects Mail List <AE-List@media-motion.tv>
Sent: Wednesday, February 20, 2019 10:59 AM
To: After Effects Mail List <AE-List@media-motion.tv>
Subject: Re: [AE] Visually lossless codec thoughts

=C2=A0

Did you try photo-jpeg in an mov? Definitely worth a= test.

=C2=A0

On Wed, Feb 20, 2019 at 10:48 AM Stephen van Vuuren = <AE-List@me= dia-motion.tv> wrote:

X264 lossless and x265 are both fails. X264 at 10-bi= t, AE says is 8-bit. X265 take 10x time to render out in Media Encoder and = takes an unknown time (at least 100x) to build a frame. I gave up after 11 hours =E2=80=93 every other codec was 14 to 22 minutes = a frame.

=C2=A0

Adobe and x265 simply don=E2=80=99t play in my exper= ience.

=C2=A0

From: After Effects Mail List <AE-List@media-motion.tv>
Sent: Wednesday, February 20, 2019 5:57 AM
To: After Effects Mail List <AE-List@media-motion.tv>
Subject: Re: [AE] Visually lossless codec thoughts

=C2=A0

x264 has lossless compression methods and can also b= e 10bit.

AfterCodec has support for 10bit and YUV 4:4:4, not = the lossless mode as far as i can see, but a high bitrate x264 should be be= tter than the old simpler DCT based Prores or DNxHR at comparable bitrates

=C2=A0

Now the question is if AE will import them correctly= again.

=C2=A0

=C2=A0

Robert

=C2=A0

=C2=A0

Den ons. 20. feb. 2019 kl. 09.08 skrev Evan Fotis Ge= orgoulakis <AE-List@media-motion.tv>:

Personally I have the opposite experience, with Pror= es producing about
20% larger files than same flavor Cineform. I tried CFYUV10bit.

On 20-Feb-19 01:07, Teddy Gage wrote:
> Chris AFAIK cineform files are even bigger than Prores at the same > flavor.


+---End of message---+
To unsubscribe send any message to <ae-list-off@media-motion.tv><= /u>


=C2=A0

--

____________________= _________

Teddy Gage

President

Shotgun Inc.

360 VR | MOGRAPH | VFX

BKLYN NY 11215

--00000000000051a03c05826c13eb--