[Support]: detections all suddenly broken on rknn #20946

derzahla · 2025-11-17T02:59:16Z

derzahla
Nov 17, 2025

Describe the problem you are having

After running an 'apt upgrade' on my OPI5+ running Armbian nobel (including a kernel upgrade) and updating Frigate to the latest dev build(from a slightly older dev build, frigate:a623150-rk) all detections are boned and I'm getting a bunch of RGA related errors.

The first thing I noticed was that the rknn runtime fails to initialize because for some reason it is trying to use /dev/dri/renderD129, which doesnt exist. I only have /dev/dri/renderD128!

2025-11-17 02:42:43.069015293  2025-11-17 02:42:43.068959669 [W:onnxruntime:Default, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card0/device/vendor"
...
frigate17  | 2025-11-17 02:37:02.749494077    import pkg_resources
frigate17  | 2025-11-17 02:37:02.749495536  W rknn-toolkit-lite2 version: 2.3.2
frigate17  | 2025-11-17 02:37:02.753914832  **************************************************
frigate17  | 2025-11-17 02:37:02.753920373  !!! It is detected that some necessary files are missing in the container.
frigate17  | 2025-11-17 02:37:02.753943123  !!! When starting the container, please use the -v parameter to map the corresponding files in the host to the container.
frigate17  | 2025-11-17 02:37:02.753944290  The reference parameters of run container are as follows:
frigate17  | 2025-11-17 02:37:02.753995039      -v /dev/dri/renderD129:/dev/dri/renderD129
frigate17  | 2025-11-17 02:37:02.754000581      -v /proc/device-tree/compatible:/proc/device-tree/compatible
frigate17  | 2025-11-17 02:37:02.754003498  **************************************************
frigate17  | 2025-11-17 02:37:02.754104705  E Catch exception when init runtime!
frigate17  | 2025-11-17 02:37:02.754902698  E Traceback (most recent call last):
frigate17  | 2025-11-17 02:37:02.754905907    File "/usr/local/lib/python3.11/dist-packages/rknnlite/api/rknn_lite.py", line 148, in init_runtime
frigate17  | 2025-11-17 02:37:02.754907657      self.rknn_runtime = RKNNRuntime(root_dir=self.root_dir, target=target, device_id=device_id,
frigate17  | 2025-11-17 02:37:02.754909115                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate17  | 2025-11-17 02:37:02.754910573    File "rknnlite/api/rknn_runtime.py", line 349, in rknnlite.api.rknn_runtime.RKNNRuntime.__init__
frigate17  | 2025-11-17 02:37:02.754914365    File "rknnlite/api/rknn_runtime.py", line 394, in rknnlite.api.rknn_runtime.RKNNRuntime._check_container
frigate17  | 2025-11-17 02:37:02.754916698  RuntimeError
frigate17  | 2025-11-17 02:37:02.754917573  
frigate17  | 2025-11-17 02:37:02.759643408  Process frigate.detector:rknn:
frigate17  | 2025-11-17 02:37:02.759647492  Traceback (most recent call last):
frigate17  | 2025-11-17 02:37:02.759648950    File "/usr/lib/python3.11/multiprocessing/process.py", line 314, in _bootstrap
frigate17  | 2025-11-17 02:37:02.759649825      self.run()
frigate17  | 2025-11-17 02:37:02.759651283    File "/opt/frigate/frigate/object_detection/base.py", line 138, in run
frigate17  | 2025-11-17 02:37:02.759652742      object_detector = LocalObjectDetector(detector_config=self.detector_config)
frigate17  | 2025-11-17 02:37:02.759654200                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate17  | 2025-11-17 02:37:02.759655658    File "/opt/frigate/frigate/object_detection/base.py", line 61, in __init__
frigate17  | 2025-11-17 02:37:02.759656825      self.detect_api = create_detector(detector_config)
frigate17  | 2025-11-17 02:37:02.759657992                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate17  | 2025-11-17 02:37:02.759698825    File "/opt/frigate/frigate/detectors/__init__.py", line 18, in create_detector
frigate17  | 2025-11-17 02:37:02.759700283      return api(detector_config)
frigate17  | 2025-11-17 02:37:02.759701450             ^^^^^^^^^^^^^^^^^^^^
frigate17  | 2025-11-17 02:37:02.759702616    File "/opt/frigate/frigate/detectors/plugins/rknn.py", line 65, in __init__
frigate17  | 2025-11-17 02:37:02.759723033      self.runner = RKNNModelRunner(
frigate17  | 2025-11-17 02:37:02.759724199                    ^^^^^^^^^^^^^^^^
frigate17  | 2025-11-17 02:37:02.759725658    File "/opt/frigate/frigate/detectors/detection_runners.py", line 427, in __init__
frigate17  | 2025-11-17 02:37:02.759726533      self._load_model()
frigate17  | 2025-11-17 02:37:02.759727991    File "/opt/frigate/frigate/detectors/detection_runners.py", line 442, in _load_model
frigate17  | 2025-11-17 02:37:02.759729158      raise RuntimeError("Failed to initialize RKNN runtime")
frigate17  | 2025-11-17 02:37:02.759730616  RuntimeError: Failed to initialize RKNN runtime

So I tried adding the mapping "/dev/dri/renderD128:/dev/dri/renderD129" to my docker-compose.yml. Now the rknn runtime starts but all detections are failing with strange errors related to RGA:

frigate17  | 2025-11-17 02:46:28.894925685  [2025-11-17 02:46:28] watchdog.litokam1              ERROR   : Ffmpeg process crashed unexpectedly for litokam1.
frigate17  | 2025-11-17 02:46:28.895860467  [2025-11-17 02:46:28] watchdog.litokam1              ERROR   : The following ffmpeg logs include the last 100 lines prior to exit.
frigate17  | 2025-11-17 02:46:28.896780375  [2025-11-17 02:46:28] ffmpeg.litokam1.detect         ERROR   : [Parsed_scale_rkrga_0 @ 0xaaab11b66790] RGA blit failed: -5
frigate17  | 2025-11-17 02:46:28.898001279  [2025-11-17 02:46:28] ffmpeg.litokam1.detect         ERROR   : [vf#1:0 @ 0xaaab1197ee50] Error while filtering: Generic error in an external library
frigate17  | 2025-11-17 02:46:28.898619606  [2025-11-17 02:46:28] ffmpeg.litokam1.detect         ERROR   : Failed to inject frame into filter network: Generic error in an external library
frigate17  | 2025-11-17 02:46:28.899369474  [2025-11-17 02:46:28] ffmpeg.litokam1.detect         ERROR   : Error while filtering: Generic error in an external library
frigate17  | 2025-11-17 02:46:28.900586295  [2025-11-17 02:46:28] ffmpeg.litokam1.detect         ERROR   : [out#0/segment @ 0xaaab1197a050] Nothing was written into output file, because at least one of its streams received no packets.
frigate17  | 2025-11-17 02:46:28.901594285  [2025-11-17 02:46:28] ffmpeg.litokam1.detect         ERROR   : [out#1/rawvideo @ 0xaaab11c20540] Nothing was written into output file, because at least one of its streams received no packets.

So I immediately downgraded the kernel packages that we upgraded and went back to the 'known good' frigate build that I was previously using (a623150-rk) and oddly enough the problem is still present! I have no idea what could have possibly happened. please haalp!

Version

a623150-rk

What browser(s) are you using?

firefox

Frigate config file

#environment_vars:
#  LIBVA_DRIVER_NAME: radeonsi
#  HSA_OVERRIDE_GFX_VERSION: "9.0.0"
logger:
  default: info
  logs:
    frigate.data_processing.real_time.face: debug
    frigate.embeddings.maintainer: debug

mqtt:
  enabled: true
  host: mqtt.domain.local
  user: frigate
  password: 

birdseye:
  enabled: true
  mode: motion

ffmpeg:
  hwaccel_args: preset-rkmpp
  input_args: -avoid_negative_ts make_zero -fflags nobuffer -fflags 
    +genpts+discardcorrupt -flags low_delay -strict experimental 
    -analyzeduration 500M -probesize 500M -rw_timeout 5000000
    #input_args: -avoid_negative_ts make_zero -fflags +genpts+discardcorrupt+nobuffer -flags low_delay -strict experimental -analyzeduration 15000000 -probesize 1000M -rw_timeout 5000000

detectors: # required
  rknn: # required
    type: rknn # required
    # number of NPU cores to use
    # 0 means choose automatically
    # increase for better performance if you have a multicore NPU e.g. set to 3 on rk3588
    num_cores: 3

model: # required
  # name of model (will be automatically downloaded) or path to your own .rknn model file
  # possible values are:
  path: frigate-fp16-yolov9-t
  model_type: yolo-generic
  width: 320
  height: 320
    #input_pixel_format: bgr # required
  # shape of detection frame
  #input_tensor: nhwc
    #input_dtype: float_denorm
    #input_dtype: float
  # needs to be adjusted to model, see below
  labelmap_path: /labelmap/coco-80.txt
  labelmap:
    36: egg

# Optional: Object configuration
# NOTE: Can be overridden at the camera level
objects:
  # Optional: list of objects to track from labelmap.txt (default: shown below)
  track:
    - person
    - car
      #- motorcycle
    - bird
    - dog
  # Optional: mask to prevent all object types from being detected in certain areas (default: no mask)
  # Checks based on the bottom center of the bounding box of the object.
  # NOTE: This mask is COMBINED with the object type specific mask below
  mask: 0.000,0.000,0.781,0.000,0.781,0.278,0.000,0.278
  # Optional: filters to reduce false positives for specific object types
  filters:
    person:
      # Optional: minimum width*height of the bounding box for the detected object (default: 0)
      min_area: 5000
      # Optional: maximum width*height of the bounding box for the detected object (default: 24000000)
      max_area: 1000000
      # Optional: minimum width/height of the bounding box for the detected object (default: 0)
      min_ratio: 0.5
      # Optional: maximum width/height of the bounding box for the detected object (default: 24000000)
      max_ratio: 2.0
      # Optional: minimum score for the object to initiate tracking (default: shown below)
      min_score: 0.7
      # Optional: minimum decimal percentage for tracked object's computed score to be considered a true positive (default: shown below)
      threshold: 0.8
      # Optional: mask to prevent this object type from being detected in certain areas (default: no mask)
      # Checks based on the bottom center of the bounding box of the object
      mask: 0.000,0.000,0.781,0.000,0.781,0.278,0.000,0.278

record:
  enabled: true
  continuous:
    days: 2
  motion:
    days: 7
  alerts:
    retain:
      days: 30
      mode: motion
  detections:
    retain:
      days: 30
      mode: motion

snapshots:
  enabled: true
  clean_copy: true
  timestamp: true
  bounding_box: true
  retain:
    default: 8

semantic_search:
  # Optional: Enable semantic search (default: shown below)
  enabled: true
  # Optional: Re-index embeddings database from historical tracked objects (default: shown below)
  reindex: false
  # Optional: Set the model used for embeddings. (default: shown below)
  model: jinav1
  # Optional: Set the model size used for embeddings. (default: shown below)
  # NOTE: small model runs on CPU and large model runs on GPU
  model_size: small

face_recognition:
  # Optional: Enable semantic search (default: shown below)
  enabled: true
  # Optional: Minimum face distance score required to save the attempt (default: shown below)
  #min_score: 0.8
  # Optional: Minimum face detection score required to detect a face (default: shown below)
  # NOTE: This only applies when not running a Frigate+ model
  detection_threshold: 0.8
  # Optional: Minimum face distance score required to be considered a match (default: shown below)
  recognition_threshold: 0.9
  # Optional: Min area of detected face box to consider running face recognition (default: shown below)
  #min_area: 1200
  min_area: 400
  # Optional: Save images of recognized faces for training (default: shown below)
  save_attempts: 200
  # Optional: Apply a blur quality filter to adjust confidence based on the blur level of the image (default: shown below)
  blur_confidence_filter: true
  model_size: large

go2rtc:
  streams:
    #  jaw10z101:
    #  - "ffmpeg:rtsp://admin:temp1234@192.168.1.235:554/live/ch00_0#audio=copy#video=h264#hardware"
    #  - "ffmpeg:jaw10z101#video=copy#audio=AAC#hardware"
    #jaw10z102:
    #  - "ffmpeg:rtsp://admin:temp1234@192.168.1.235:554/live/ch01_0#audio=copy#video=h264#hardware"
    #  - "ffmpeg:jaw10z102#video=copy#audio=AAC#hardware"
    wvdb1:
      # - exec:/usr/lib/ffmpeg/6.0/bin/ffmpeg -hide_banner -v error 
      #  -allowed_media_types video+audio -fflags nobuffer -flags low_delay 
      #  -fflags +genpts+discardcorrupt -hwaccel rkmpp -timeout 5000000 
      #  -user_agent go2rtc/ffmpeg -strict experimental -i 
      #  rtsp://thingino:bASSWORDz@wvdb1.iot.domain.local:554/ch0 -c copy -user_agent
      #  ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {{output}}
      - ffmpeg:rtsp://thingino:bASSWORDz@wvdb1.iot.domain.local:554/ch0#video=h264#rotate=90#width=1080#height=1920#audio=copy#hardware
      - ffmpeg:wvdb1#audio=aac#video=copy#hardware
        #- ffmpeg:wvdb1#audio=copy#video=h264#hardware
        #- ffmpeg:wvdb1#audio=aac#video=h264#hardware
        #- ffmpeg:wvdb1#audio=aac
        #    wvdb1_ch1:
        #    - "ffmpeg:wvdb1#audio=copy#video=h264#width=640#height=360#hardware"
        #- "ffmpeg:wvdb1_ch1#audio=aac#video=copy"
    litokam1:
      #- exec:/usr/lib/ffmpeg/6.0/bin/ffmpeg -hide_banner -v error 
      #  -allowed_media_types video+audio -fflags nobuffer -flags low_delay 
      #  -fflags +genpts+discardcorrupt -hwaccel rkmpp -timeout 5000000 
      #  -user_agent go2rtc/ffmpeg -strict experimental -i 
      #  rtsp://thingino:bASSWORDz@litokam1.iot.domain.local:554/ch0 -c copy 
      #  -user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {{output}}
      - ffmpeg:rtsp://thingino:bASSWORDz@litokam1.iot.domain.local:554/ch0#video=h264#audio=copy#hardware
      - ffmpeg:litokam1#audio=aac#video=copy#hardware

    amcrest:
      - ffmpeg:rtsp://camuser:BASSWORDz@amcrest.iot.domain.local:554/cam/realmonitor?channel=1&subtype=0
      - ffmpeg:amcrest#audio=opus#video=copy#hardware
    ginzzu1:
      #- exec:/usr/lib/ffmpeg/6.0/bin/ffmpeg -hide_banner -v error 
      #  -allowed_media_types video+audio -fflags nobuffer -flags low_delay 
      #  -fflags +genpts+discardcorrupt -hwaccel rkmpp -timeout 5000000 
      #  -user_agent go2rtc/ffmpeg -strict experimental -i 
      #  rtsp://thingino:bASSWORDz@ginzzu1.iot.domain.local:554/ch0 -c copy 
      #  -user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {{output}}
      - ffmpeg:rtsp://thingino:bASSWORDz@ginzzu1.iot.domain.local:554/ch0#video=h264#audio=copy#hardware
      - ffmpeg:ginzzu1#audio=aac#video=copy#hardware
        #ginzzu2:
        # - exec:/usr/lib/ffmpeg/6.0/bin/ffmpeg -hide_banner -v error 
        #  -allowed_media_types video+audio -fflags nobuffer -flags low_delay 
        #  -fflags +genpts+discardcorrupt -hwaccel rkmpp -timeout 5000000 
        #  -user_agent go2rtc/ffmpeg -strict experimental -i 
        # rtsp://thingino:bASSWORDz@ginzzu2.iot.domain.local:554/ch0 -c copy 
        #-user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {{output}}
        #  - ffmpeg:rtsp://thingino:bASSWORDz@ginzzu2.iot.domain.local:554/ch0#input=rtsp/udp
        #- ffmpeg:ginzzu2#audio=aac
    cinnado-d1:
      - ffmpeg:rtsp://thingino:bASSWORDz@cinnado-d1.iot.domain.local:554/ch0
      - ffmpeg:cinnado-d1#audio=aac#video=copy#hardware
    cinnado-d3:
      - ffmpeg:rtsp://thingino:bASSWORDz@cinnado-d3.iot.domain.local:554/ch0
      - ffmpeg:cinnado-d3#audio=aac#video=copy#hardware
    webrtc:
      candidates:
        - frigate.domain.local:8555
        - stun:8555

telemetry:
  # Optional: Enabled network interfaces for bandwidth stats monitoring (default: empty list, let nethogs search all)
  network_interfaces:
    - eth0
  # Optional: Configure system stats
  stats:
    # Enable network bandwidth stats monitoring for camera ffmpeg processes, go2rtc, and object detectors. (default: shown below)
    # NOTE: The container must either be privileged or have cap_net_admin, cap_net_raw capabilities enabled.
    network_bandwidth: true
  # Optional: Enable the latest version outbound check (default: shown below)
  # NOTE: If you use the HomeAssistant integration, disabling this will prevent it from reporting new versions
  version_check: true

cameras:
  jaw10z101:
    enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/jaw10z101
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
            - detect
    onvif:
      host: 192.168.1.235
      port: 8899
      user: admin
      password: temp1234
      ignore_time_mismatch: true
      autotracking:
        enabled: false
        calibrate_on_startup: false
        zooming: absolute
        return_preset: Home
        timeout: 30
        # A lower value will keep more of the scene in view around a tracked object.
        # A higher value will zoom in more on a tracked object, but Frigate may lose tracking more quickly.
        # The value should be between 0.1 and 0.75
        track:
          - person
          - car
          - motorcycle
          - license_plate
        required_zones:
          - default
        movement_weights: 1.0, 1.0, 0.07363510131835938, 0.11022404752751835, 
          0.19117384033818394, 0
    detect:
      enabled: false
      width: 2304
      height: 1296
      fps: 5
    objects:
      track:
        - person
        - car
      mask: 0,0,0.622,0,0.6,0.099,0,0.241
      filters:
        person:
          mask:
            - 0,0,0.781,0,0.366,0.132,0,0.278
            - 0,0,0.622,0,0.6,0.099,0,0.241
    audio:
      enabled: true
    record:
      enabled: true
    motion:
      threshold: 80
      contour_area: 10
      improve_contrast: true
    zones:
      Default:
        coordinates: 
          0.084,0.278,0.646,0.21,1,0.334,1,0.705,0.01,0.731,0.001,0.657,0,0.541,0.003,0.317
        inertia: 2
        loitering_time: 0
        objects:
          - car
          - person
      road:
        coordinates: 0.61,0.134,0.617,0.069,0.042,0.171,0.054,0.293
        loitering_time: 0

  jaw10z102:
    enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/jaw10z102
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
            - detect
    detect:
      enabled: false
      width: 2304
      height: 1296
      fps: 5
    objects:
      track:
        - person
        - car
      mask: 0,0,0.622,0,0.6,0.099,0,0.241
      filters:
        person:
          mask:
            - 0,0,0.781,0,0.366,0.132,0,0.278
            - 0,0,0.622,0,0.6,0.099,0,0.241
    audio:
      enabled: true
    record:
      enabled: true
    motion:
      threshold: 80
      contour_area: 10
      improve_contrast: true
    zones:
      Default:
        coordinates: 
          0.084,0.278,0.646,0.21,1,0.334,1,0.705,0.01,0.731,0.001,0.657,0,0.541,0.003,0.317
        inertia: 2
        loitering_time: 0
        objects:
          - car
          - person
      road:
        coordinates: 0.61,0.134,0.617,0.069,0.042,0.171,0.054,0.293
        loitering_time: 0
  litokam1:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/litokam1
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
            - detect
              #     - path: rtsp://thingino:bASSWORDz@litokam1.iot.domain.local:554/ch1
              #   roles:
    onvif:
      host: litokam1.iot.domain.local
      port: 80
      user: thingino
      password: bASSWORDz
      autotracking:
        enabled: false
        calibrate_on_startup: false
        zooming: absolute
        return_preset: Home
        timeout: 30
        # A lower value will keep more of the scene in view around a tracked object.
        # A higher value will zoom in more on a tracked object, but Frigate may lose tracking more quickly.
        # The value should be between 0.1 and 0.75
        track:
          - person
          - car
          - motorcycle
          - license_plate
        required_zones:
          - default
        movement_weights: 1.0, 1.0, 0.07363510131835938, 0.11022404752751835, 
          0.19117384033818394, 0
    detect:
      enabled: true
      width: 2304
      height: 1296
      fps: 5
    objects:
      track:
        - person
        - car
      mask: 0,0,0.622,0,0.6,0.099,0,0.241
      filters:
        person:
          mask:
            - 0,0,0.781,0,0.366,0.132,0,0.278
            - 0,0,0.622,0,0.6,0.099,0,0.241
    audio:
      enabled: true
    record:
      enabled: true
    motion:
      threshold: 80
      contour_area: 10
      improve_contrast: true
    zones:
      Default:
        coordinates: 
          0.084,0.278,0.646,0.21,1,0.334,1,0.705,0.01,0.731,0.001,0.657,0,0.541,0.003,0.317
        inertia: 2
        loitering_time: 0
        objects:
          - car
          - person
      road:
        coordinates: 0.61,0.134,0.617,0.069,0.042,0.171,0.054,0.293
        loitering_time: 0
        objects: car
    review: {}
  amcrest:
    enabled: true
    lpr:
      enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/amcrest
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
            - detect
    onvif:
      host: amcrest.iot.domain.local
      port: 80
      user: admin
      password: BASSWORDz
      autotracking:
        enabled: false
        calibrate_on_startup: true
        zooming: disabled
          #zoom_factor: 0.3
        track:
          - person
        return_preset: '1'
        timeout: 5
        required_zones:
          - default
    detect:
      enabled: true
      fps: 5
    objects:
      track:
        - person
    zones:
      default:
        coordinates: 0.041,0,0.516,0.01,0.969,0.282,1,1,0.137,1
        loitering_time: 0
        inertia: 2
        objects: person

  cinnado-d1:
    enabled: true
    lpr:
      enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/cinnado-d1
          input_args: preset-rtsp-restream
          roles:
            - record
 #       - path: rtsp://thingino:bASSWORDz@cinnado-d1.iot.domain.local:554/ch1
 #        hwaccel_args: preset-rk-h264
 #         roles:
            - audio
            - detect
    onvif:
      host: cinnado-d1.iot.domain.local
      port: 80
      user: thingino
      password: bASSWORDz
      autotracking:
        enabled: false
        calibrate_on_startup: false
        zooming: disabled
          #zoom_factor: 0.3
        track:
          - person
        return_preset: Home
        timeout: 5
        required_zones:
          - default
    detect:
      enabled: true
      fps: 4
      width: 1920
      height: 1080
    objects:
      track:
        - person
    audio:
      enabled: true
    record:
      enabled: true
    motion:
      threshold: 50
      contour_area: 30
      improve_contrast: true
    zones:
      default:
        coordinates: 0.041,0,0.516,0.01,0.969,0.282,1,1,0.137,1
        loitering_time: 0
        inertia: 2
        objects: person
    review:
      alerts:
        required_zones: default
      detections:
        required_zones: default
  cinnado-d3:
    enabled: true
    lpr:
      enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/cinnado-d3
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
            - detect
    onvif:
      host: cinnado-d3.iot.domain.local
      port: 80
      user: thingino
      password: bASSWORDz
      autotracking:
        enabled: false
        calibrate_on_startup: false
        zooming: disabled
          #zoom_factor: 0.3
        track:
          - person
        return_preset: Home
        timeout: 10
        required_zones:
          - default
    detect:
      enabled: true
      fps: 4
      width: 1280
      height: 720
    objects:
      track:
        - person
    audio:
      enabled: true
    record:
      enabled: true
    motion:
      threshold: 40
      contour_area: 20
      improve_contrast: true
    zones:
      default:
        coordinates: 0,0,0,0.933,1,0.861,1,0
        loitering_time: 0
        inertia: 3
    review:
      alerts:
        required_zones: default
      detections:
        required_zones: default
  wvdb1:
    enabled: true
    lpr:
      enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/wvdb1
          input_args: preset-rtsp-restream
          roles:
            - record
              #        - path: rtsp://127.0.0.1:8554/wvdb1_ch1
              #          input_args: preset-rtsp-restream
              #          roles:
            - detect
            - audio
        #   - path: rtsp://thingino:bASSWORDz@wvdb1.iot.domain.local:554/ch1
    detect:
      enabled: true
      width: 720
      height: 1280
      fps: 5
    objects:
      track:
        - person
        - car
          # - motorcycle
    audio:
      enabled: true
    record:
      enabled: true
    motion:
      threshold: 25
      contour_area: 10
      improve_contrast: true
    zones:
      Default:
        coordinates: 
          0.662,0.813,0.639,0.694,0.73,0.623,1,0.56,1,0.23,0.602,0.217,0.256,0.305,0,0.474,0,0.668,0,0.879,0.004,0.922,0,1,1,1,1,0.792,0.805,0.731
        loitering_time: 0
        inertia: 3
        objects: person
    review:
      detections:
        required_zones: Default
  ginzzu1:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/ginzzu1
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
     #   - path: rtsp://thingino:bASSWORDz@ginzzu1.iot.domain.local:554/ch1
          #hwaccel_args: preset-rk-h264
     #     roles:
            - detect
    onvif:
      host: ginzzu1.iot.domain.local
      port: 80
      user: thingino
      password: bASSWORDz
      autotracking:
        enabled: false
        calibrate_on_startup: false
        zooming: disabled
        zoom_factor: 0.3
        track:
          - person
          - car
          - motorcycle
          - license_plate
            #          - dog
            #     - bird
        return_preset: Home
        timeout: 10
        required_zones:
          - default
    detect:
      enabled: true
      fps: 3
      width: 1920
      height: 1080
    objects:
      track:
        - person
          #   - animal
          #      - dog
          #    - bird
        - car
        - motorcycle
    audio:
      enabled: true
    record:
      enabled: true
    motion:
      threshold: 40
      contour_area: 40
      improve_contrast: true
    zones:
      default:
        coordinates: 1,1,0.983,0.08,0.237,0,0,0.024,0.003,0.995
        loitering_time: 0
        inertia: 3
    review:
      alerts:
        required_zones: default
      detections:
        required_zones: default
  ginzzu2:
    enabled: false
    lpr:
      enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/ginzzu2
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
            - detect
              #      - path: rtsp://thingino:bASSWORDz@ginzzu2.iot.domain.local:554/ch1
            #          roles:
            #            - audio
            #            - detect
    onvif:
      host: ginzzu2.iot.domain.local
      port: 80
      user: thingino
      password: bASSWORDz
      autotracking:
        enabled: false
        calibrate_on_startup: false
        zooming: disabled
          #zoom_factor: 0.3
        track:
          #  - animal
          #    - dog
          - bird
          - egg
        return_preset: Home
        timeout: 5
        required_zones:
          - default
    detect:
      enabled: true
      width: 640
      height: 360
      fps: 1
    objects:
      track:
        #    - animal
        #     - dog
        - bird
        - egg
    audio:
      enabled: true
    record:
      enabled: true
    motion:
      threshold: 30
      contour_area: 70
      improve_contrast: true
    zones:
      default:
        coordinates: 0.99,0.984,0.867,0.348,0.237,0,0,0.024,0.003,0.995
        loitering_time: 0
version: 0.17-0
camera_groups:
  Home:
    order: 1
    icon: LuHome
    cameras:
      - birdseye
      - litokam1
      - cinnado-d1
      - cinnado-d2
      - cinnado-d3
      - wvdb1
      - ginzzu1
      - ginzzu2
  Birdseye:
    order: 2
    icon: LuAlignEndVertical
    cameras: birdseye
detect:
  enabled: true
lpr:
  enabled: false
classification:
  #  custom:
  #  deer:
  #    threshold: 0.8
  #    object_config:
  #      objects: [deer] # object labels to classify
  #      classification_type: sub_label # or: attribute
  #  chicken:
  #    threshold: 0.8
  #    object_config:
  #      objects: [chicken] # object labels to classify
  #      classification_type: sub_label # or: attribute
  bird:
    enabled: false

Relevant Frigate log output

added above

Relevant go2rtc log output

added above

FFprobe output from your camera

Input #0, rtsp, from 'rtsp://thingino:caldjbckds@litokam1.iot.domain.local:554/ch0':
  Metadata:
    title           : thingino prudynt
    comment         : stream0
  Duration: N/A, start: 0.000188, bitrate: N/A
  Stream #0:0: Video: hevc (Main), yuv420p(tv, bt709), 2304x1296, 10 fps, 10 tbr, 90k tbn
  Stream #0:1: Audio: opus, 48000 Hz, stereo, fltp
  Stream #0:2: Audio: aac, 16000 Hz, mono, fltp
  Stream #0:3: Audio: pcm_mulaw, 8000 Hz, mono, s16, 64 kb/s
  Stream #0:4: Audio: pcm_alaw, 8000 Hz, mono, s16, 64 kb/s

Frigate stats

No response

Install method

Docker Compose

docker-compose file or Docker CLI command

services:
  frigate:
    container_name: frigate17
    restart: unless-stopped
      #image: ghcr.io/blakeblackshear/frigate:ea576e7-rk
      #image: ghcr.io/blakeblackshear/frigate:4f4c8a4-rk
      #image: ghcr.io/blakeblackshear/frigate:e9dc302-rk 
      #image: ghcr.io/blakeblackshear/frigate:bdb7a18-rk
      #image: ghcr.io/blakeblackshear/frigate:923412e-rk
      #image: ghcr.io/blakeblackshear/frigate:1f061a8-rk
      #image: ghcr.io/blakeblackshear/frigate:4bea695-rk
      #image: ghcr.io/blakeblackshear/frigate:33f0c23-rk
      #image: ghcr.io/blakeblackshear/frigate:b1a5896-rk
      #image: ghcr.io/blakeblackshear/frigate:6caa205-rk
      #image: ghcr.io/blakeblackshear/frigate:4228861-rk
      #image: ghcr.io/blakeblackshear/frigate:c5fec32-rk
      #image: ghcr.io/blakeblackshear/frigate:d6f5d2b-rk
      #image: ghcr.io/blakeblackshear/frigate:6ccf8cd-rk
      #image: ghcr.io/blakeblackshear/frigate:62bc2ae-rk
      #image: ghcr.io/blakeblackshear/frigate:d44340e-rk
      #image: ghcr.io/blakeblackshear/frigate:85f7138-rk
      #image: ghcr.io/blakeblackshear/frigate:e4eac4a-rk
      #image: ghcr.io/blakeblackshear/frigate:f1a05d0-rk
      #image: ghcr.io/blakeblackshear/frigate:097673b-rk
    image: ghcr.io/blakeblackshear/frigate:a623150-rk







    shm_size: "1500mb"
    privileged: true
    volumes:
      - /sys/:/sys/:ro
      - ./config:/config
      - ./storage:/media/frigate
        # - ./overrides/go2rtc:/usr/local/go2rtc/bin/go2rtc
        #   - ./overrides/ffmpeg-cur/ffplay:/usr/lib/ffmpeg/6.0/bin/ffplay
        # - ./overrides/ffmpeg-cur/ffmpeg:/usr/lib/ffmpeg/6.0/bin/ffmpeg
        # - ./overrides/ffmpeg-cur/ffprobe:/usr/lib/ffmpeg/6.0/bin/ffprobe
      - /dev/dri/renderD128:/dev/dri/renderD129
      - /proc/device-tree/compatible:/proc/device-tree/compatible
      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "1984:1984"
      - "5000:5000"
      - "8971:8971"
      - "8554:8554" # RTSP feeds
      - "8555:8555" # RTSP feeds
    security_opt:
      - apparmor=unconfined
      - systempaths=unconfined
    devices:
      - /dev/dri/
      - /dev/dma_heap/
      - /dev/rga
      - /dev/mpp_service

Object Detector

RKNN

Network connection

Wired

Camera make and model

various

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

No response

derzahla · 2025-11-17T02:59:51Z

inkeep-ai-assistant[bot]
Bot Nov 17, 2025

I see several issues with your setup that are likely causing the RGA errors and detection failures.

Device Mapping Issue

The RKNN runtime is looking for /dev/dri/renderD129 because that's what the runtime expects, but your system only has /dev/dri/renderD128. Your workaround of mapping renderD128:renderD129 is not the correct solution(1).

According to the Rockchip installation documentation, you should be mapping the devices correctly. Remove the line:

- /dev/dri/renderD128:/dev/dri/renderD129

And ensure you have the proper device mappings as documented(1):

devices:
  - /dev/dri
  - /dev/dma_heap
  - /dev/rga
  - /dev/mpp_service

You already have these in your compose file, which is correct.

Privileged Mode

Your container is running in privileged mode (privileged: true), which gives it access to all devices. This might be causing conflicts with your explicit device mappings(2). Try removing privileged: true and rely only on the explicit device mappings and security options you've already configured(1):

security_opt:
  - apparmor=unconfined
  - systempaths=unconfined

RGA Filter Errors

The RGA blit errors ([Parsed_scale_rkrga_0 @ 0xaaab11b66790] RGA blit failed: -5) suggest that the hardware acceleration pipeline is failing. This is likely related to the incorrect device mapping or driver issues after your kernel upgrade.

Recommended Steps

Remove the privileged: true line from your docker-compose.yml
Remove the /dev/dri/renderD128:/dev/dri/renderD129 mapping
Keep your existing device mappings as they are (they're correct per the docs)
Verify your kernel has the proper Rockchip drivers loaded after the upgrade:

ls /dev/dri

You should see renderD128 and renderD129 (the NPU device)(1).

Check that the RKNPU driver is properly loaded:

sudo cat /sys/kernel/debug/rknpu/version

If the kernel upgrade broke the RKNPU drivers, you may need to reinstall them following the Rockchip installation prerequisites(1).

📚 Sources:

Have another question? Just tag @inkeep.

2 replies

derzahla Nov 17, 2025
Author

Ok so I must have had a /dev/renderD129 before then. Now I don't. wtf. But it shows the RKNPU driver is loaded.

$ sudo cat /sys/kernel/debug/rknpu/version
RKNPU driver: v0.9.8
$ sudo ls /dev/dri
by-path  card0	renderD128

derzahla Nov 17, 2025
Author

@inkeep I tried going back to armbian linux-image-vendor-rk35xx:arm64 (25.5.1) kernel. still no /dev/dri/renderD129. Finally I tried going back to Joshua Riek's linux-image-6.1.0-1026-rockchip - and now /dev/dri/renderD129 shows up! I dont get it, I had been running fine on the standard armbian linux-image-vendor-rk35xx kernel for 2 weeks since switching to armbian. What could explain why now I am unable to get /dev/dri/renderD129 to show up with the armbian vendor kernels?

2025-11-17T04:08:48Z

inkeep-ai-assistant[bot]
Bot Nov 17, 2025

The issue you're experiencing is related to kernel-level driver support for the Rockchip NPU, not specifically a Frigate configuration problem.

What's Happening

The /dev/dri/renderD129 device represents the NPU (Neural Processing Unit) on your Rockchip platform(1). When this device is missing, it means the RKNPU driver isn't creating the device node properly, even though the driver itself reports as loaded.

Why the Armbian Vendor Kernel Lost renderD129

Based on the Rockchip installation prerequisites, Frigate requires "a linux distribution that comes with the rockchip BSP kernel 5.10 or 6.1 and necessary drivers (especially rkvdec2 and rknpu)"(1). The documentation specifically recommends checking:

$ ls /dev/dri
by-path card0 card1 renderD128 renderD129 # should list renderD128 (VPU) and renderD129 (NPU)

(1)

The fact that Joshua Riek's linux-image-6.1.0-1026-rockchip kernel works while the Armbian vendor kernel doesn't suggests:

Driver Configuration Differences: The Armbian vendor kernel may have been updated with different RKNPU driver configurations or module parameters that prevent the NPU device node from being created properly
Kernel Module Loading Order: The newer Armbian kernel might load modules in a different sequence, causing the NPU initialization to fail silently
BSP Patches: Joshua Riek's kernel likely includes specific Rockchip BSP (Board Support Package) patches that the Armbian vendor kernel is missing or has modified

Why It Worked Before

You mentioned it worked for 2 weeks on the Armbian vendor kernel. This suggests either:

A kernel update within those 2 weeks changed something
The apt upgrade you ran updated kernel modules or firmware that affected NPU initialization
A configuration file or device tree overlay was modified during the upgrade

Recommendation

Since Joshua Riek's kernel works reliably, stick with it for now. The Armbian vendor kernel appears to have incomplete or misconfigured RKNPU support. You could:

Continue using Joshua Riek's linux-image-6.1.0-1026-rockchip kernel
Report this regression to the Armbian project, as their vendor kernel should properly support the NPU
Check Armbian forums to see if others have reported similar NPU device node issues after recent updates

The good news is that with the working kernel, your Frigate setup should function properly with the correct device mappings as documented(1).

📚 Sources:

Installation

Have another question? Just tag @inkeep.

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Support]: detections all suddenly broken on rknn #20946

Uh oh!

{{title}}

Uh oh!

Replies: 2 comments 2 replies

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{editor}}'s edit

{{editor}}'s edit

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Select a reply

Uh oh!

Uh oh!

[Support]: detections all suddenly broken on rknn #20946

Uh oh!

derzahla Nov 17, 2025

Describe the problem you are having

Version

What browser(s) are you using?

Frigate config file

Relevant Frigate log output

Relevant go2rtc log output

FFprobe output from your camera

Frigate stats

Install method

docker-compose file or Docker CLI command

Object Detector

Network connection

Camera make and model

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

Replies: 2 comments · 2 replies

Uh oh!

inkeep-ai-assistant[bot] Bot Nov 17, 2025

Device Mapping Issue

Privileged Mode

RGA Filter Errors

Recommended Steps

Uh oh!

Uh oh!

derzahla Nov 17, 2025 Author

Uh oh!

derzahla Nov 17, 2025 Author

Uh oh!

inkeep-ai-assistant[bot] Bot Nov 17, 2025

What's Happening

Why the Armbian Vendor Kernel Lost renderD129

Why It Worked Before

Recommendation

derzahla
Nov 17, 2025

Replies: 2 comments 2 replies

inkeep-ai-assistant[bot]
Bot Nov 17, 2025

derzahla Nov 17, 2025
Author

derzahla Nov 17, 2025
Author

inkeep-ai-assistant[bot]
Bot Nov 17, 2025