* update cereal
* run but not use
* log distraction type
* regression scaling
* clean up naming
* add calib buf
* add to header
* fake model
* no calib model
* adjust threshs
* 018a305f
* fix bn
* tweak1
* tweak2
* 0ff2/666
* tweak3
* t4
* t5
* fix out of bound
* skip when replaying old segments
* update ref
* fix onnxmodel
* get calib
* update model replay refs
* up ref
* mse err from 0.028070712 -> 5.8073703e-05
* build with weights fixup
* need thneed lib also
* don't break for binaries
* static analysis says i need init
* check the bias
* load_dlc_weights
* nicer scons
* tested scons
* fix static
* pylint issue
* new ref
* a few more asserts
Co-authored-by: Harald Schafer <harald.the.engineer@gmail.com>
* Fix bigmodel on tici: cam intrinsics arent static
* Give TICI route own name for import
* Make function useable in offline test
* typo
* change tici ecam focal to trained focal
* No shadow variable
* Fixed ref
* Force update this time
* Update model replay ref commit
* one conv with defines
* add conv
* building works on C3
* this is num_outputs too, process replay is so useful
Co-authored-by: Comma Device <device@comma.ai>
* Added wide cam vipc client and bigmodel transform logic
* Added wide_frame to ModelState, should still work normally
* Refactored image input into addImage method, should still work normally
* Updated thneed/compile.cc
* Bigmodel, untested: 44f83118-b375-4d4c-ae12-2017124f0cf4/200
* Have to initialize extra buffer in SNPEModel
* Default paramater value in the wrong place I think
* Move USE_EXTRA to SConscript
* New model: 6c34d59a-acc3-4877-84bd-904c10745ba6/250
* move use extra check to runtime, not on C2
* this is always true
* more C2 checks
* log if frames are out of sync
* more logging on no frame
* store in pointer
* print sof
* add sync logic
* log based on sof difference as well
* keep both models
* less assumptions
* define above thneed
* typo
* simplify
* no need for second client is main is already wide
* more comments update
* no optional reference
* more logging to debug lags
* add to release files
* both defines
* New model: 6831a77f-2574-4bfb-8077-79b0972a2771/950
* Path offset no longer relevant
* Remove duplicate execute
* Moved bigmodel back to big_supercombo.dlc
* add wide vipc stream
* Tici must be tici
* Needs state too
* add wide cam support to model replay
* handle syncing better
* ugh, c2
* print that
* handle ecam lag
* skip first one
* so close
* update refs
Co-authored-by: mitchellgoffpc <mitchellgoffpc@gmail.com>
Co-authored-by: Harald Schafer <harald.the.engineer@gmail.com>
Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
Co-authored-by: Comma Device <device@comma.ai>
* add thneed optimizer
* local work group opt
* kernels and final mods
* release files
* build system touchups
* fix kernel path, rand inputs for self test
* broken since extra is gone
* update model replay ref
Co-authored-by: Comma Device <device@comma.ai>
* Convert all text strings to f-strings
Reformats all the text from the old "%-formatted" and .format(...) format to the newer f-string format, as defined in PEP 498. This requires Python 3.6+.
Flynt 0.69 was used to reformat the strings. 120 f-strings were created in 51 files.
F-strings are in general more readable, concise and performant. See also: https://www.python.org/dev/peps/pep-0498/#rationale
* revert pyextra changes
* revert ublox.py
Co-authored-by: Willem Melching <willem.melching@gmail.com>
* model output struct for metadata
* better max element search
* ModelDataRawDesireProb.to_array()
* eliminate some copies
* not worth messing with softmax now
* remove unused includes
* more cleanup
* no longer pointers
* better variable name
* fix recurrent state
* improve variable name
* fix OUTPUT_SIZE and NET_OUTPUT_SIZE
* Replace lists with generators v2
* Replace set with {}
Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
* Replace more set() with {}
Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
* lane lines and road edges
* roll back some changes to find what broke things
* rollback more changes to find issue
* fix order of lane line probs
* try outputs on stack again
* initialize output array
* model pose more struct
* constexpr seems useful
* more concise
* fix order
* seems more readable
* plan struct
* attempt to fix compiling on device
* fix constexpr errors on device
* fix rotation rate in log
* maybe this fixes compiling constexpr on device
* exp() cannot be used in constexpr on c2
* only copy what we actually log
* simplify pivot()
* fix setting position std
* see if this works
* try lambda again
* giving up on the lambda to select member
* fix position std
* fix param order
* c3 is straightforward
* c2 transform
* c3 preview
* alpha model
* Revert "alpha model": wrong quantization data
This reverts commit b50e573973.
* yuv looks good
* a55b
* const int
* block top deadzone
* new box alpha curve
* update k
* fix debug
* rhd use same offset
* box shift for rhd
* 7e3e
* set new thresh
* update ref
* explain
* remove e2e
* same metric thresh
* process isocc
* model
* cancel extra policy
* update numbers
* new model
* update cereal
* up cereal
Co-authored-by: Comma Device <device@comma.ai>
* debug commit
* cleanup
* some indexing bugs
* need more its
* BALANCE is way better it seems
* fix test
* this converges in 2000segs
* new ref
* less cpu
* 6be443f2-ed70-4580-9c31-61b94d068e24/950
* meta changed too
* new ref
* new ref from CI
* unused
* Revert "new ref from CI"
This reverts commit 28b6bdc810.
* new ref from CI again
ZwX1616