[Support Dev]: License Plate model not working after running the latest dev version

[weitheng]

Describe the problem you are having

LPR not working after using the latest dev version a7bbca5, getting the following error. Face Recognition works fine

Version

a7bbca5

What browser(s) are you using?

No response

Frigate config file

detectors:
  onnx:
    type: onnx

model:
  model_type: rfdetr
  width: 320
  height: 320
  input_tensor: nchw
  input_dtype: float
  path: /config/rfdetr-Medium.onnx

detect:
  fps: 5
  enabled: true
  max_disappeared: 25
  stationary:
    interval: 50
    threshold: 50

lpr:
  enabled: true
  device: GPU
  model_size: small
  min_area: 600
  detection_threshold: 0.50
  recognition_threshold: 0.94
  enhancement: 2
  min_plate_length: 3
  match_distance: 1

Relevant Frigate log output

2025-09-21 23:45:51.048977951 [E:onnxruntime:, sequential_executor.cc:572 ExecuteKernel] Non-zero status code returned while running NonMaxSuppression node. Name:'/end2end/NonMaxSuppression' Status Message: std::bad_alloc: cudaErrorStreamCaptureUnsupported: operation not permitted when stream is capturing


Error running YOLOv9 license plate detection model: Error in execution: /onnxruntime_src/onnxruntime/core/providers/cuda/cuda_call.cc:129 std::conditional_t<THRW, void, onnxruntime::common::Status> onnxruntime::CudaCall(ERRTYPE, const char*, const char*, SUCCTYPE, const char*, const char*, int) [with ERRTYPE = cudaError; bool THRW = true; SUCCTYPE = cudaError; std::conditional_t<THRW, void, common::Status> = void] /onnxruntime_src/onnxruntime/core/providers/cuda/cuda_call.cc:121 std::conditional_t<THRW, void, onnxruntime::common::Status> onnxruntime::CudaCall(ERRTYPE, const char*, const char*, SUCCTYPE, const char*, const char*, int) [with ERRTYPE = cudaError; bool THRW = true; SUCCTYPE = cudaError; std::conditional_t<THRW, void, common::Status> = void] CUDA failure 901: operation failed due to a previous error during capture ; GPU=0 ; hostname=ubuntu-frigate ; file=/onnxruntime_src/onnxruntime/core/providers/cuda/cuda_graph.cc ; line=61 ; expr=cudaStreamEndCapture(stream_, &graph);

Relevant go2rtc log output

N/A

FFprobe output from your camera

N/A

Frigate stats

No response

Install method

Docker Compose

docker-compose file or Docker CLI command

services:
  frigate:
    container_name: frigate
    privileged: true # this may not be necessary for all setups
    image: ghcr.io/blakeblackshear/frigate:a7bbca5-tensorrt
    shm_size: "3g"
    restart: unless-stopped
    devices:
      - /dev/bus/usb:/dev/bus/usb
    runtime: nvidia
    volumes:
      - "/root/frigate/config/:/config/"
      - /etc/timezone:/etc/timezone:ro
      - /etc/localtime:/etc/localtime:ro
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 2000000000
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
          memory: 12288M
          cpus: '10.00'
        limits:
          memory: 13312M
          cpus: '13.00'
    ports:
      - '5000:5000'
      - '8971:8971'
      - '8081:8080'
      - "8554:8554"  # RTSP Feeds
      - "8555:8555"
    environment:
      NVIDIA_DRIVER_CAPABILITIES: 'all'
      USE_FP16: false
    network_mode: host

Object Detector

TensorRT

Network connection

Wired

Camera make and model

Dahua, Hikvision

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

No response

[NickM-27]

Thanks, I'll take a look

[NickM-27]

It would be odd, but possible that the license plate yolov9 model has something different that is not compatible with CUDA graphs

[NickM-27]

Fixed in #20159

[weitheng]

Thanks for the quick fix!