[HW Accel Support]: AttributeError: 'TensorRtDetector' object has no attribute 'outputs'

[cdgriffith]

Describe the problem you are having

On a TrueNAS Scale instance I am trying to set up a frigate instance with nvidia passthrough with a 3060 RTX (Have also tried Nvidia L4 and Nvidia 1080).

In a single instance, I am able to build the model, but then it fails to use it.

TrueNAS Scale docker config:

If I change the detector to cpu from tensorrt firgate lists the 3060 on the webpage, and will even use the 3060 for ffmpeg encodings.

Other's with the issue seem to have mount path issues (not applicable here as it's in a single instance) or forget to rebuild between card switches, which is also not issue seen here to my knowledge.

Version

stable-tensorrt

Frigate config file

mqtt:
  enabled: false

detectors:
  detector_name:
    type: tensorrt

cameras:
  front:
    enabled: true
    ffmpeg:
      hwaccel_args: preset-nvidia-h265
      inputs:
        - path: rtsp://...
          roles:
            - detect
        - path: rtsp://...
          roles:
            - record
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5

  driveway:
    enabled: true
    ffmpeg:
      hwaccel_args: preset-nvidia-h265    
      inputs:
        - path: rtsp://...
          roles:
            - detect        
        - path: rtsp://...
          roles:
            - record
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5

  backyard:
    enabled: true
    ffmpeg:
      hwaccel_args: preset-nvidia-h265    
      inputs:
        - path: rtsp://...
          roles:
            - detect
        - path: rtsp://...
          roles:
            - record
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5
   
  garden:
    enabled: true
    ffmpeg:
      hwaccel_args: preset-nvidia-h265    
      inputs:
        - path: rtsp://...
          roles:
            - detect
        - path: rtsp://...
          roles:
            - record
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5  
      
  catroom:
    enabled: true
    ffmpeg:
      hwaccel_args: preset-nvidia-h264   
      inputs:
        - path: rtsp://...
          roles:
            - detect
        - path: rtsp://...
          roles:
            - record
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5
      

auth:
  reset_admin_password: false
version: 0.14

record:
  enabled: true
  events:
    pre_capture: 10
    post_capture: 10
  retain:
    mode: all

snapshots:
  enabled: true

docker-compose file or Docker CLI command

N/A

Relevant Frigate log output

2024-12-12 03:54:17.339306+00:00Creating yolov7-320.cfg and yolov7-320.weights
2024-12-12 03:54:17.342354+00:002024-12-12T03:54:17.342354857Z
2024-12-12 03:54:17.342368+00:00Done.
2024-12-12 03:54:21.149334+00:002024-12-12T03:54:21.149334902Z
2024-12-12 03:54:21.149375+00:00Generating yolov7-320.trt. This may take a few minutes.
2024-12-12 03:54:21.149403+00:002024-12-12T03:54:21.149403971Z
2024-12-12 03:54:22.875875+00:00/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py:149: DeprecationWarning: Use network created with NetworkDefinitionCreationFlag::EXPLICIT_BATCH flag instead.
2024-12-12 03:54:22.875972+00:00builder.max_batch_size = MAX_BATCH_SIZE
2024-12-12 03:54:22.876041+00:00/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py:151: DeprecationWarning: Use set_memory_pool_limit instead.
2024-12-12 03:54:22.876089+00:00config.max_workspace_size = 1 << 30
2024-12-12 03:54:22.876099+00:00/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py:174: DeprecationWarning: Use build_serialized_network instead.
2024-12-12 03:54:22.876108+00:00engine = builder.build_engine(network, config)
2024-12-12 03:56:27.014371+00:00Generated yolov7-320.trt in 126 seconds
2024-12-12 03:56:27.014414+00:00Available tensorrt models:
2024-12-12 03:56:27.015493+00:00yolov7-320.trt
2024-12-12 03:56:27.016099+00:00s6-rc: info: service trt-model-prepare successfully started
2024-12-12 03:56:27.016230+00:00s6-rc: info: service frigate: starting
2024-12-12 03:56:27.018017+00:00s6-rc: info: service frigate successfully started
2024-12-12 03:56:27.018126+00:00s6-rc: info: service nginx: starting
...
2024-12-12 03:56:30.919941+00:002024-12-12 03:56:30.919882654  Process detector:detector_name:
2024-12-12 03:56:30.920489+00:002024-12-12 03:56:30.920457370  Traceback (most recent call last):
2024-12-12 03:56:30.920514+00:002024-12-12 03:56:30.920475313    File "/usr/lib/python3.9/multiprocessing/process.py", line 315, in _bootstrap
2024-12-12 03:56:30.920545+00:002024-12-12 03:56:30.920476355      self.run()
2024-12-12 03:56:30.920560+00:002024-12-12 03:56:30.920477057    File "/usr/lib/python3.9/multiprocessing/process.py", line 108, in run
2024-12-12 03:56:30.920571+00:002024-12-12 03:56:30.920477628      self._target(*self._args, **self._kwargs)
2024-12-12 03:56:30.920594+00:002024-12-12 03:56:30.920485522    File "/opt/frigate/frigate/object_detection.py", line 102, in run_detector
2024-12-12 03:56:30.920603+00:002024-12-12 03:56:30.920486113      object_detector = LocalObjectDetector(detector_config=detector_config)
2024-12-12 03:56:30.920611+00:002024-12-12 03:56:30.920486675    File "/opt/frigate/frigate/object_detection.py", line 53, in __init__
2024-12-12 03:56:30.920633+00:002024-12-12 03:56:30.920487185      self.detect_api = create_detector(detector_config)
2024-12-12 03:56:30.920642+00:002024-12-12 03:56:30.920487716    File "/opt/frigate/frigate/detectors/__init__.py", line 18, in create_detector
2024-12-12 03:56:30.920651+00:002024-12-12 03:56:30.920491373      return api(detector_config)
2024-12-12 03:56:30.920660+00:002024-12-12 03:56:30.920491934    File "/opt/frigate/frigate/detectors/plugins/tensorrt.py", line 246, in __init__
2024-12-12 03:56:30.920684+00:002024-12-12 03:56:30.920503035      self.engine = self._load_engine(detector_config.model.path)
2024-12-12 03:56:30.920694+00:002024-12-12 03:56:30.920504277    File "/opt/frigate/frigate/detectors/plugins/tensorrt.py", line 90, in _load_engine
2024-12-12 03:56:30.920714+00:002024-12-12 03:56:30.920519566      with open(model_path, "rb") as f, trt.Runtime(self.trt_logger) as runtime:
2024-12-12 03:56:30.920723+00:002024-12-12 03:56:30.920525828  TypeError: expected str, bytes or os.PathLike object, not NoneType
2024-12-12 03:56:30.920732+00:002024-12-12 03:56:30.920526399  Exception ignored in: <function TensorRtDetector.__del__ at 0x7fce6115a430>
2024-12-12 03:56:30.920748+00:002024-12-12 03:56:30.920526890  Traceback (most recent call last):
2024-12-12 03:56:30.920757+00:002024-12-12 03:56:30.920527421    File "/opt/frigate/frigate/detectors/plugins/tensorrt.py", line 265, in __del__
2024-12-12 03:56:30.920767+00:002024-12-12 03:56:30.920604575      if self.outputs is not None:
2024-12-12 03:56:30.920776+00:002024-12-12 03:56:30.920610206  AttributeError: 'TensorRtDetector' object has no attribute 'outputs'

Relevant go2rtc log output

N/A

FFprobe output from your camera

N/A

Install method

Docker Compose

Object Detector

TensorRT

Network connection

Wired

Camera make and model

hikvision 4K, amcrest 4K, reolink 1440p

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[NickM-27]

Please read the docs on configuring a tensorrt model https://docs.frigate.video/configuration/object_detectors#configuration-parameters

[basjes1977]

To make @NickM-27 his answer more clear,
you are missing the model bit:

model:
path: /config/model_cache/tensorrt/yolov7-320.trt
input_tensor: nchw
input_pixel_format: rgb
width: 320
height: 320

I ran into exactly the same issue. For me this was what was missing and caused exactly your error message. So hopefully this helps you too.