mirror of
https://github.com/sunnypilot/sunnypilot.git
synced 2026-08-21 09:13:46 +08:00
George Hotz
29cd51b876
* compiling, won't work yet
* running with inputs and outputs
* there's some magic chance this works
* no more dlc, include onnx
* yolo tests plz
* bump tinygrad
* files_common + delete dlc
* tinygrad_repo -> tinygrad
* pre commit config
* llops needed
* extra in files_common
* bump tinygrad
* fix indent
* tinygrad/nn/__init__
* tinygrad_repo
* bump tinygrad repo
* bump tinygrad
* bump with native_exp, match maybe
* native_explog is argument
* pyopencl no cache
* 5% chance this matches
* work in float32?
* bump tinygrad
* fix build
* no __init__
* fix recip
* dumb hack
* adding thneed PC support
* fix pc segfault
* pc thneed is working
* to_image
* prints stuff with debug=2
* it sort of works
* copy host ptr is simpler
* bug fix
* build on c3
* this correct?
* reenable float16
* fix private, fixup copy_inputs internal
* bump tinygrad and update ref commit
* fix OPTWG on PC
* maybe fix non determinism
* revert model replay ref commit
* comments, init zeroed out buffers
* upd ref commit
* bump tinygrad to fix initial image
* try this ref
Co-authored-by: Comma Device <device@comma.ai>
old-commit-hash: 40d6f4b65c
Neural networks in openpilot
To view the architecture of the ONNX networks, you can use netron
Supercombo
Supercombo input format (Full size: 393738 x float32)
- image stream
- Two consecutive images (256 * 512 * 3 in RGB) recorded at 20 Hz : 393216 = 2 * 6 * 128 * 256
- Each 256 * 512 image is represented in YUV420 with 6 channels : 6 * 128 * 256
- Channels 0,1,2,3 represent the full-res Y channel and are represented in numpy as Y[::2, ::2], Y[::2, 1::2], Y[1::2, ::2], and Y[1::2, 1::2]
- Channel 4 represents the half-res U channel
- Channel 5 represents the half-res V channel
- Each 256 * 512 image is represented in YUV420 with 6 channels : 6 * 128 * 256
- Two consecutive images (256 * 512 * 3 in RGB) recorded at 20 Hz : 393216 = 2 * 6 * 128 * 256
- wide image stream
- Two consecutive images (256 * 512 * 3 in RGB) recorded at 20 Hz : 393216 = 2 * 6 * 128 * 256
- Each 256 * 512 image is represented in YUV420 with 6 channels : 6 * 128 * 256
- Channels 0,1,2,3 represent the full-res Y channel and are represented in numpy as Y[::2, ::2], Y[::2, 1::2], Y[1::2, ::2], and Y[1::2, 1::2]
- Channel 4 represents the half-res U channel
- Channel 5 represents the half-res V channel
- Each 256 * 512 image is represented in YUV420 with 6 channels : 6 * 128 * 256
- Two consecutive images (256 * 512 * 3 in RGB) recorded at 20 Hz : 393216 = 2 * 6 * 128 * 256
- desire
- one-hot encoded vector to command model to execute certain actions, bit only needs to be sent for 1 frame : 8
- traffic convention
- one-hot encoded vector to tell model whether traffic is right-hand or left-hand traffic : 2
- recurrent state
- The recurrent state vector that is fed back into the GRU for temporal context : 512
Supercombo output format (Full size: XXX x float32)
Read here for more.
Driver Monitoring Model
- .onnx model can be run with onnx runtimes
- .dlc file is a pre-quantized model and only runs on qualcomm DSPs
input format
- single image (640 * 320 * 3 in RGB):
- full input size is 6 * 640/2 * 320/2 = 307200
- represented in YUV420 with 6 channels:
- Channels 0,1,2,3 represent the full-res Y channel and are represented in numpy as Y[::2, ::2], Y[::2, 1::2], Y[1::2, ::2], and Y[1::2, 1::2]
- Channel 4 represents the half-res U channel
- Channel 5 represents the half-res V channel
- normalized, ranging from -1.0 to 1.0
output format
- 39 x float32 outputs (parsing example)
- face pose: 12 = 6 + 6
- face orientation [pitch, yaw, roll] in camera frame: 3
- face position [dx, dy] relative to image center: 2
- normalized face size: 1
- standard deviations for above outputs: 6
- face visible probability: 1
- eyes: 20 = (8 + 1) + (8 + 1) + 1 + 1
- eye position and size, and their standard deviations: 8
- eye visible probability: 1
- eye closed probability: 1
- wearing sunglasses probability: 1
- poor camera vision probability: 1
- face partially out-of-frame probability: 1
- (deprecated) distracted probabilities: 2
- face covered probability: 1
- face pose: 12 = 6 + 6