StarPilot/openpilot/selfdrive/modeld/dmonitoringmodeld.py

#!/usr/bin/env python3
import os
import base64
from openpilot.starpilot.models.runtime_bridge.runtime import prepare as prepare_model_runtime
prepare_model_runtime('monitoring')
from openpilot.selfdrive.modeld.helpers import MODELS_DIR, load_oob
from tinygrad.tensor import Tensor
import time
import pickle
import numpy as np

from openpilot.cereal import messaging
from openpilot.cereal.messaging import PubMaster, SubMaster
from openpilot.cereal.visionipc import VisionStreamType
from msgq.visionipc import VisionIpcClient, VisionBuf
from openpilot.common.swaglog import cloudlog
from openpilot.common.realtime import config_realtime_process
from openpilot.common.transformations.model import dmonitoringmodel_intrinsics
from openpilot.common.transformations.camera import _ar_ox_fisheye, _os_fisheye
from openpilot.system.camerad.cameras.nv12_info import get_nv12_info
from openpilot.selfdrive.modeld.parse_model_outputs import sigmoid, safe_exp

SEND_RAW_PRED = os.getenv('SEND_RAW_PRED')
MODEL_PKL_PATH = MODELS_DIR / 'dmonitoring_model_tinygrad.pkl'


class ModelState:
  inputs: dict[str, np.ndarray]
  output: np.ndarray

  def __init__(self, cam_w: int, cam_h: int):
    jits = load_oob(MODEL_PKL_PATH)
    self.DEV = jits['input_specs']['input_img'][2]
    self.input_shapes = jits['metadata']['input_shapes']
    self.output_slices = pickle.loads(base64.b64decode(jits['metadata']['metadata']['output_slices']))

    self.numpy_inputs = {
      'calib': np.zeros(self.input_shapes['calib'], dtype=np.float32),
    }

    self.warp_inputs_np = {'transform': np.zeros((3,3), dtype=np.float32)}
    self.warp_inputs = {k: Tensor(v, device='NPY') for k,v in self.warp_inputs_np.items()}
    self.frame_buf_params = get_nv12_info(cam_w, cam_h)
    self.tensor_inputs = {k: Tensor(v, device=self.DEV).realize() for k,v in self.numpy_inputs.items()}
    self.calib_host = Tensor(self.numpy_inputs['calib'], device='NPY')._buffer()
    self._blob_cache : dict[int, Tensor] = {}
    self.model_run = jits['run']
    self.outputs = {name: Tensor(np.zeros(shape, dtype=dtype), device=device).realize() for name, (shape, dtype, device) in jits['output_specs'].items()}
    warp_path = MODELS_DIR / f'dm_warp_{cam_w}x{cam_h}_tinygrad.pkl'
    from openpilot.starpilot.models.runtime_bridge.runtime import selected_legacy, load_warp
    if selected_legacy():
      self.image_warp = load_warp(warp_path)
    else:
      with warp_path.open('rb') as f:
        self.image_warp = pickle.load(f)['run']

  def run(self, buf: VisionBuf, calib: np.ndarray, transform: np.ndarray) -> tuple[np.ndarray, float]:
    self.numpy_inputs['calib'][0,:] = calib

    t1 = time.perf_counter()
    self.tensor_inputs['calib']._buffer().copy_from(self.calib_host)

    ptr = np.frombuffer(buf.data, dtype=np.uint8).ctypes.data
    # There is a ringbuffer of imgs, just cache tensors pointing to all of them
    if ptr not in self._blob_cache:
      self._blob_cache[ptr] = Tensor.from_blob(ptr, (self.frame_buf_params[3],), dtype='uint8', device=self.DEV)

    self.warp_inputs_np['transform'][:] = transform[:]
    self.tensor_inputs['input_img'] = self.image_warp(input_frame=self._blob_cache[ptr], M_inv=self.warp_inputs['transform'])

    self.model_run(output_buffers=self.outputs, **self.tensor_inputs)
    output = self.outputs['outputs'].numpy().astype(np.float32).reshape(-1)

    t2 = time.perf_counter()
    return output, t2 - t1

def slice_outputs(model_outputs, output_slices):
  return  {k: model_outputs[np.newaxis, v] for k,v in output_slices.items()}

def parse_model_output(model_output):
  parsed = {}
  parsed['wheel_on_right'] = sigmoid(model_output['wheel_on_right'])
  for ds_suffix in ['lhd', 'rhd']:
    face_descs = model_output[f'face_descs_{ds_suffix}']
    parsed[f'face_descs_{ds_suffix}'] = face_descs[:, :-6]
    parsed[f'face_descs_{ds_suffix}_std'] = safe_exp(face_descs[:, -6:])
    for key in ['face_prob', 'left_eye_prob', 'right_eye_prob','left_blink_prob', 'right_blink_prob', 'sunglasses_prob', 'using_phone_prob', 'sleep_prob']:
      parsed[f'{key}_{ds_suffix}'] = sigmoid(model_output[f'{key}_{ds_suffix}'])
  return parsed

def fill_driver_data(msg, model_output, ds_suffix):
  msg.faceOrientation = model_output[f'face_descs_{ds_suffix}'][0, :3].tolist()
  msg.faceOrientationStd = model_output[f'face_descs_{ds_suffix}_std'][0, :3].tolist()
  msg.facePosition = model_output[f'face_descs_{ds_suffix}'][0, 3:5].tolist()
  msg.facePositionStd = model_output[f'face_descs_{ds_suffix}_std'][0, 3:5].tolist()
  msg.faceProb = model_output[f'face_prob_{ds_suffix}'][0, 0].item()
  msg.leftEyeProb = model_output[f'left_eye_prob_{ds_suffix}'][0, 0].item()
  msg.rightEyeProb = model_output[f'right_eye_prob_{ds_suffix}'][0, 0].item()
  msg.leftBlinkProb = model_output[f'left_blink_prob_{ds_suffix}'][0, 0].item()
  msg.rightBlinkProb = model_output[f'right_blink_prob_{ds_suffix}'][0, 0].item()
  msg.sunglassesProb = model_output[f'sunglasses_prob_{ds_suffix}'][0, 0].item()
  msg.phoneProb = model_output[f'using_phone_prob_{ds_suffix}'][0, 0].item()
  msg.sleepProb = model_output[f'sleep_prob_{ds_suffix}'][0, 0].item()

def get_driverstate_packet(model_output, frame_id: int, location_ts: int, exec_time: float, gpu_exec_time: float):
  msg = messaging.new_message('driverStateV2', valid=True)
  ds = msg.driverStateV2
  ds.frameId = frame_id
  ds.modelExecutionTime = exec_time
  ds.gpuExecutionTime = gpu_exec_time
  ds.rawPredictions = model_output['raw_pred']
  ds.wheelOnRightProb = model_output['wheel_on_right'][0, 0].item()
  fill_driver_data(ds.leftDriverData, model_output, 'lhd')
  fill_driver_data(ds.rightDriverData, model_output, 'rhd')
  return msg


def main():
  config_realtime_process(7, 5)

  cloudlog.warning("connecting to cabin stream")
  vipc_client = VisionIpcClient("camerad", VisionStreamType.VISION_STREAM_CABIN, True)
  while not vipc_client.connect(False):
    time.sleep(0.1)
  assert vipc_client.is_connected()
  cloudlog.warning(f"connected with buffer size: {vipc_client.buffer_len}")

  model = ModelState(vipc_client.width, vipc_client.height)
  cloudlog.warning("models loaded, dmonitoringmodeld starting")

  sm = SubMaster(["extrinsicsCalibration"])
  pm = PubMaster(["driverStateV2"])

  calib = np.zeros(model.numpy_inputs['calib'].size, dtype=np.float32)
  model_transform = None

  while True:
    buf = vipc_client.recv()
    if buf is None:
      continue

    if model_transform is None:
      cam = _os_fisheye if buf.width == _os_fisheye.width else _ar_ox_fisheye
      model_transform = np.linalg.inv(np.dot(dmonitoringmodel_intrinsics, np.linalg.inv(cam.intrinsics))).astype(np.float32)

    sm.update(0)
    if sm.updated["extrinsicsCalibration"]:
      calib[:] = np.array(sm["extrinsicsCalibration"].rpyCalib)

    t1 = time.perf_counter()
    model_output, gpu_execution_time = model.run(buf, calib, model_transform)
    t2 = time.perf_counter()
    raw_pred = model_output.tobytes() if SEND_RAW_PRED else b''
    model_output = slice_outputs(model_output, model.output_slices)
    model_output = parse_model_output(model_output)
    model_output['raw_pred'] = raw_pred
    msg = get_driverstate_packet(model_output, vipc_client.frame_id, vipc_client.timestamp_sof, t2 - t1, gpu_execution_time)
    pm.send("driverStateV2", msg)


if __name__ == "__main__":
  try:
    main()
  except KeyboardInterrupt:
    cloudlog.warning("got SIGINT")