* it's something
* backup
* 16:10
* cleanup
* this is fine
* close
* remove some junk
* no heck
* disos
* real 10
* for some reason this is flipped
* 20hz
* no return
* ae
* tear
* need curve laster
* correct real gains
* fix time
* cleanup
* why the scam
* disable for now
* 0.7
* hdr
* that doesnt work
* what
* hugeoof
* clean up
* cleanup
* fix regs
* welp cant
* is this corrent
* it is sq
* remove
* back
* stg10bit
* back2ten
* Revert "remove"
This reverts commit 18712ab7e1.
* 20hz and swb
* correct height
* 10bit
* ui hack for now
* slight
* perfect
* blk64
* ccm
* fix page faults
* template
* set 4x
* is this fine
* try
* this seems to work
* Revert "this seems to work"
This reverts commit d3c9023d3f.
* needs to be static
* close
* 64 is optimal
* 2
* take
* not 1
* offset
* whats going on
* i have no idea
* less resistence
* box defs
* no
* reduce blur artifacts
* simplify
* fix
* fake short is too much for bright
* can be subzero
* should not use lsvc
* no wasted bit
* cont no slow
* no less than 10bit
* it is based
* wrong
* right
* quart
* shift
* raise noise floor
* 4.5/4.7
* same ballpark
* int is fine
* shane owes me m4a4
* Revert "shane owes me m4a4"
This reverts commit b4283fee18.
* back
* Revert "4.5/4.7"
This reverts commit e38f96e90c.
* default
* oof
* clean up
* simpilfy
* from sensorinfo
* no div
* better name
* not the wrong one
* not anymore relevant
* too
* not call it debayer
* cl headers
* arg is 2nd
* gone is is_bggr
* define
* no is hdr
* rgb_tmp
* p1
* clean up
* 4
* cant for
* fix somewhre else
* const
* ap
* rects
* just set staruc
* nnew tmp
* pull it for now
* 12
* common rect
* Revert "not anymore relevant"
This reverts commit 1d574673a1.
* Revert "too"
This reverts commit c2d4dcc52a.
* Revert "Revert "too""
This reverts commit 0abbabe1fd.
* no tol is fine
* rename
* sensor id
* unsgin
* flag
* some linalg
* cast
* should be h ref
* cap
---------
Co-authored-by: Comma Device <device@comma.ai>
* it's something
* backup
* 16:10
* cleanup
* this is fine
* close
* remove some junk
* no heck
* disos
* real 10
* for some reason this is flipped
* 20hz
* no return
* ae
* tear
* need curve laster
* correct real gains
* fix time
* cleanup
* why the scam
* disable for now
* 0.7
* hdr
* that doesnt work
* what
* hugeoof
* clean up
* cleanup
* fix regs
* welp cant
* is this corrent
* it is sq
* remove
* back
* add base comment
* clean up
* make orders clear
* not abcd
---------
Co-authored-by: Comma Device <device@comma.ai>
* start porting OS04C10
* something
* is the pll why this doesn't work
* understand plls, still doesn't work
* work
* weee something
* working?
* timing
* encoderd works with less cameras
* 1928
* flip image
* color balance
* adjust black level
* extra height maybe
* cleanups
* clean up sensor i2c
* sensors
* approx 20fps
* remove mirror
* 50ms, no fsin yet
* remove imx390 support
* some attempt at AE
* linearize the response
* OX changes (#25697)
* organize
* wb
* best effort but probably messed up by artifacts
* multicam sync
Co-authored-by: Comma Device <device@comma.ai>
* clean up
* more
* fix multiexposure
* start porting OS04C10
* something
* is the pll why this doesn't work
* understand plls, still doesn't work
* work
* weee something
* working?
* timing
* encoderd works with less cameras
* 1928
* flip image
* color balance
* adjust black level
* extra height maybe
* cleanups
* clean up sensor i2c
* sensors
* approx 20fps
* remove mirror
* 50ms, no fsin yet
* remove imx390 support
* some attempt at AE
* linearize the response
* OX changes (#25697)
* organize
* wb
* best effort but probably messed up by artifacts
* multicam sync
Co-authored-by: Comma Device <device@comma.ai>
* clean up
* more
* fix multiexposure
* unbroken AE
* oops
* sort of good AE
* both should work 1
* open then init
* fix debayer
* rename
* fix test build
Co-authored-by: Comma Device <device@comma.ai>
Co-authored-by: ZwX1616 <zwx1616@gmail.com>
Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
* rgb to nv12
* nv12 works (w memcpy)
* correct now
* no copy
* fix nv12 with fast debayer
* reverts of unused stuff
* ui use nv12
* comment out thumbnails for now
* rebase fix
* dm read nv12
* model read nv12
* fix ffmpeg encoder
* thumbnails from nv12
* replay to nv12
* python framereader support nv12
* remove hardcoded frame/buffer sizes
* fix build
* ffmpeg encoder fix buffers
* small cleanup
* reduce power usage test
* fix cpu usage test
* fix snapshot
* fix loggerd test
* bump cereal
Co-authored-by: Comma Device <device@comma.ai>
Co-authored-by: Joost Wooning <jwooning@gmail.com>
* remove local caching
* remove local caching
* reduce camerad power
* break stupid imx390 black level support
* handle edges
* now 13ms, vignetting is 'slightly' less correct
* halfs->floats and inlines, down to 12.9ms
* oops, fix float
* val from 12 is ushort
* don't decide vignetting in the debayer kernel
* 7.77 ms
* adding back black level support was free
* Revert "adding back black level support was free"
This reverts commit a841d17727.
* minor
* rip out unused gain and black level, remove print
* save 150mW
* fix replay test
* fix top/bottom rows
* lame left right edge fix
Co-authored-by: Comma Device <device@comma.ai>
Co-authored-by: Joost Wooning <jwooning@gmail.com>
* camerad: combine debayering and rgb_to_yuv opencl kernels
* fix border
* fix snapshot
* rename function
* update camerad cpu usage
* update camerad power draw
* vignetting equal to previously
* test other local worksize
* use less floats
* reduce amount of code
* move barrier back
* make faster
* fix corners
* cleanup
* cleanup
* allow more jitter on driverState timing
Co-authored-by: Comma Device <device@comma.ai>
* something is output
* min stuff
* visible picture
* pics look good
* maybe
* whole sensor
* fix all cameras
* support both cameras
* autoexposure for imx390
* fix imx390 blacklevel
* touchups
* put gain in db scale
* inline and fix max exposure
Co-authored-by: Comma Device <device@comma.ai>
* AE improvements
* boardd: also use gain to compute IR brightness
* use default rect for DM autoexposure
* more smooth
* whitespace
* camerad cpu usage
* hcg on is slightly better for noise
* hysteris around high conversion gain
* improve i2c timing
* use AB contexts
* HCG default off
* run at 10 hz
* stay in context A for now
* remove ae thread
* wait till next frame is started before sending i2c
* back at 20fps
* add comment and filter across 3 evs
* remove context switch code
* tuning
* recomened without HCG is 0.8x
* simplify histogram calculation
* add debug prints
* faster AE
* revert that
* lower log level
* use minimum recomended gains from datasheet
* use define
* fix min gain
* dynamic grey target
* fix build
* add low pass
* more stable rounding
* increase max gain
* always enable HCG
* correctly set min ev
* fix analog gain
* remove unused variable
* fix build
* move recomended gain into cost function
* tweak cost function a bit
* small LPF on desired ev
* fix typo
* lower ts
* clean up defines
* that was used in common
* add cast
* less cost is fine
* fix tests
* whitespace
* filtering is broken
* put back
* clip gain changes instead
* cost function tweaking
* Only go below recomended gain when absolutely necessary
* small penalty on changing gains
* protect highlights
* fix modeld scale factor
* rewrite logic for accurate and smooth AE
* fix condition
* tune speed
* move to qcom2
* not exactly 2x
* avoid hunting due to ettr policy
* tune to new curve
* gain kickdowns
* log params
* 4x legroom
* no limit needed here
* simple lowpass
* tweak
* pin to a remote core
* speedup AE and lower night exp
* this curve makes more sense
* tuning from fleet data
* fix ui
* only qcom2
Co-authored-by: Comma Device <device@comma.ai>
* nerf highlights
* prioritize less noise
* no overhead
* make focus easier
* cant finetune under high gain
* add some comments
* inspired by xx mindset
Co-authored-by: Comma Device <device@comma.ai>
ZwX1616
George Hotz
Adeeb Shihadeh
Comma Device
Joost Wooning
Willem Melching
robbederks