Current Server Time: 23:36 (Central Europe)

#apertus IRC Channel Logs

2022/11/10

Timezone: UTC


08:00
fredy
left the channel
08:00
intrac
left the channel
08:00
dos
left the channel
08:00
se6astian
left the channel
08:00
fredy
joined the channel
08:01
intrac
joined the channel
08:02
dos
joined the channel
08:02
se6astian
joined the channel
08:05
paulk
Bertl: sure :) it's an ongoing project and will probably take a while to get a prototype but will get there hopefully!
08:09
paulk
vup: rk3399 is 1080p@30, allwinner v3 is 1080p@60
08:09
paulk
but newer generations do around 4k@30
08:10
paulk
RK3588 does h.264/h.265 8k@30
08:10
paulk
it's a dedicated hw block
08:10
paulk
rockchip uses the verisilicon hantro family
08:11
paulk
for allwinner it's less clear, could be some imgtech derivative
09:09
se6astian
hi paulk, very interesting project, is your focus on higher performance than raspberry pi + hq camera or do you just enjoy working on it yourself
09:10
se6astian
there is the effort to squeeze out the max possible performance from the pi and use it as cinema camera
09:10
se6astian
https://forums.raspberrypi.com/viewtopic.php?t=296776&sid=d4197f8a2e8ab8ebe6792ddc497b2ad6&start=150
09:10
se6astian
but the hardware is of course rather limited
09:10
se6astian
but very affordable in return
09:20
paulk
se6astian: yeah it's a rather similar goal, I would just like to avoid the raspberrypi because it's not all that free software friendly
09:21
se6astian
Right
09:21
paulk
ideally I'd like to find a better sensor (at least global shutter)
14:18
davidak[m]
<paulk> "RK3588 does h.264/h.265 8k@30" <- paulk: do these encoders support ALL-I (frames) 4:2:2? that is the next best thing when ProRes 4:2:2 is not available in the camera. when it has clean feed out HDMI you can use an external recorder that support ProRes (i think all do)
14:18
davidak[m]
not sure if that makes sense with affordable sensors
14:20
paulk
davidak[m]: pretty sure you can get all-I (it's really up to the driver, also there's no firmware intermediate to prevent that)
14:20
paulk
I haven't checked 4:2:2 but that feels plausible
14:20
paulk
who knows, maybe even 4:4:4 :)
14:21
davidak[m]
sounds very good
14:21
paulk
yeah basically the gap is closing
14:22
paulk
i.e. it becomes viable for more serious things than your average user's 30fps 1080p at 4:2:0
14:23
paulk
about hdmi: "TMDS Scrambler to enable support for 2160p@60 Hz with RGB/YCbCr4:4:4 or YCbCr4:2:2"
14:23
paulk
that's for rk3588
14:23
paulk
(HDMI 2.0)
14:24
paulk
RK3399 is only YUV420 though
14:26
paulk
but well, my use case for now is to have something somewhat comparable to my Canon EOS 80D, which does 1080p 4:2:0 H.264 I/P
14:26
paulk
so nothing too fancy, but newer platforms like rk3588 do bring it further
14:27
davidak[m]
would magic latern help?
14:28
paulk
on the 80D side probably (support is early work in progress)
14:28
davidak[m]
ok
14:28
paulk
for supporting SoCs with Linux probably not
14:28
paulk
AFAIK magic lantern is very canon-centric
14:28
davidak[m]
yes
14:31
paulk
for now my production ambition is more short videos for the web (youtube channel) so I don't really need to be able to produce a high-quality DCP or such
14:45
anuejn
paulk: sounds interesting :)
14:46
anuejn
maybe one could also do something like log encoding the images with a shader before encoding them
14:46
anuejn
that could also help quality-wise
14:47
anuejn
ef lenses sound a bit to long for these tiny sensors?
14:48
anuejn
also maybe something like encoding all r, g, and b sensels seperately could be interesting
14:48
anuejn
(idk if it helps quality-wise)
14:49
paulk
anuejn: the ISP hardware typically also has gamma LUTs
14:49
paulk
although if the sensor is only linear, it might help with compression but won't really give a better dynamic range
14:50
paulk
ah yes for the EF lens that's my worry
14:50
paulk
I've made a few tries with C-mount cameras that have small sensors
14:50
paulk
and it's difficult to get focus
14:50
paulk
that's why I'm interested in large sensors too :)
16:16
anuejn
sounds interesting
16:17
anuejn
but for driving the af in a meaningful way you would also need some information
16:17
anuejn
an i fear that the rpi camera sensor does not have phase detection pixels :(
20:01
vup
I mean you don't need phase detection if you have the magic of AI :P
20:08
aombk
joined the channel
20:08
aombk2
left the channel
20:25
paulk
anuejn: yeah the isp gives sharpness information stats
20:25
paulk
but probably not phase detection indeed
20:35
anuejn
paulk: but contrast based af sucks since you dont know in which direction to drive ;P
20:35
anuejn
especially for video
21:14
vup
af for video in general usually sucks :P
21:28
paulk
yes yes
21:29
paulk
what I have in mind about it is more to draw the indication on-screen and have remote focus pulling
21:30
paulk
hopefully the lens protocol can allow reliably driving the focus coil without an external stepper
21:31
paulk
actually one the most recent canon lens I have it seems that the ring is not mechanically driving focus when in manual mode but behaves like a rotary encoder fed to the canon chip as input
21:32
paulk
which then drives the focus coil accordingly
21:32
paulk
or maybe it doesn't even go out of the lens at all
21:33
paulk
anyway not really counting on af
21:34
anuejn
yeah that would be very interesting
21:34
anuejn
how well remote focus pulling with ef lenses feels
21:34
anuejn
since they cant be actuated in an absolute manner
21:39
paulk
yeah, but at least it would allow controlling the step sensitivity
21:39
paulk
feels like it could be tweaked to be usable
21:40
paulk
I'm more worried about roundtrip time from moving the rotary encoder to getting the updated sharpness feedback on-screen
21:40
paulk
especially if doing remote as in actually wireless
21:41
paulk
but definitely worth a try
23:45
Spirit532
left the channel
23:45
Spirit532
joined the channel
23:48
aombk
left the channel
23:52
aombk
joined the channel