* catch exception outside loop
* print decompress error
* add test case for corrupt log
* fix decompressbz2 stuck if log is corrupt
* recovered from corrupt data
* add output
* ass space
* std::endl
* override load(), load from buffer
* override FrameReader::load to load from the buffer
* replace NULL with nullptr
* fix test case for corrupt log
* Trigger Build
* check bzerror too
Co-authored-by: Willem Melching <willem.melching@gmail.com>
* vipc snapshot
* front and rear need to go
* remove SEND_*
* cleanup
* put that back
* fix duplicate code
* dont start camerad on pc
* fix rgb stride
Co-authored-by: Comma Device <device@comma.ai>
Co-authored-by: Willem Melching <willem.melching@gmail.com>
* no-cache mode
* fix test cases build error
* space
* don't create cache dir in no-cache mode
* fix errors in test cases
* no_local_cache_
* set the number of connections by chunk_size
* use size_t instead of int64_t
* add test case for no-cache mode
* rename variables
* fix SIGSEGV
* cleanup
* faster decompressBZ2
* always decompress bz2
* add test cases
* prepare for python interface
* fix test cases build error
* continue
* camera_replay: cache remote file
* protected inheritance
* single option name
* TODO
* test_case for LogReader&FrameReader
* fix wrong require
* test case for FileReader
* cleanup test
* test:fix wrong filename
* check cached file's checksum
* fix mkdir permissions err
cleanup filereader
* remove initialize libav network libraries.
dd
* abort all loading if one failed
* cleanup tests
* use threadpool to limit concurrent downloads
* cache more segments
* merge 3 segments for replay
* one segment uses about 100M of memory
* use segments_need_merge.size()
* shutdown
* fix stuck if exit replay before keyboard thread started
* load one segment at a time
* small cleanup
* cleanup filereader
* space
* tiny cleanup
* merge master
* cleanup test cases
* use util:create_directories
* cleanup framereader
* new class Debayer
* merge master
* pass in cl_event
* needs define, class is not the same
* match clEnqueueCopyBuffer api
Co-authored-by: Willem Melching <willem.melching@gmail.com>
* decode from the previous keyframe after seek
* less memory
* some stream seems to contian no keyframes
* test random seek
* merge master
* continue
update test_cases
use fr
* merge master
* new function device_acquire & device_config
* add space
* apply reviews
* use the specified struct for each command
* fix typo
* apply review
* remove static
* revert device_control
* trigger rotate in eoncode thread
* rotate in time
* lgtm
* dcam trigger rotate on C3
* check trigger rotate field
* Use >=
* add rotator thread
* set cnt to 0
* test encodeIdx is increasing across segments
* test both segmentId and encodeId
* fix encodeIdx
* no thread needed
* add log for failed to encode
* stricter C3 test, dont check start encodeId on C2 dcam
* only update last_camera_seen_tms when considered for rotate
Co-authored-by: deanlee <deanlee3@gmail.com>
* 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
Willem Melching