[Detector Support]: I can't enable Alerts and Detections in version 0.17 on the Orin Nano.

[ThieryPeleias]

Describe the problem you are having

Hi Folks
Sorry if I didn't understand the system in version 0.17.

I had a PC with an RTX 3060 running version 0.16 and everything worked perfectly.

During this update, I switched to NVIDIA Jetson Orin Nano Super Developer Kit.

And I can't get the autodetect and detections to work with images. Only with audio. Even when restarting, the Detect appears as disabled and I need to manually activate it. And there are no person detections, only audio.

Also, the GPU usage seems to be low as well.

On my other PC everything worked, including Ollama. It was running smoothly, I spent several months fine-tuning it. But with Orin, or version 0.17, I can't make any progress. I've tried several AIs and I can't get anywhere.
I didn't find anything in the logs.

Thank you for your reply.

Version

0.17

Frigate config file

go2rtc:
  streams:
    entrada: 
      rtsp://camera_entrada:{FRIGATE_RTSP_PASSWORD}@192.168.255.115/stream1#rtsp_transport=tcp#video=copy
    entrada_sub: 
      rtsp://camera_entrada:{FRIGATE_RTSP_PASSWORD}@192.168.255.115/stream2#rtsp_transport=tcp#video=copy
  rtsp:
    listen: :8554

ffmpeg:
  hwaccel_args: preset-jetson-h264  # Global fallback
  retry_interval: 10
  output_args:
    detect: -threads 2 -f rawvideo -pix_fmt yuv420p
    record: preset-record-generic-audio-aac

detectors:
  tensorrt:
    type: tensorrt
    device: 0

model:
  path: /config/model_cache/tensorrt/8.6.2/yolov7-320.trt
  labelmap_path: /labelmap/coco-80.txt  # Default COCO labels; customize if needed
  input_tensor: nchw
  input_pixel_format: rgb
  width: 320
  height: 320

objects:
  track:
    - person
    - car
    - dog
    - cat
    - bird
    - bicycle
  filters:
    person:
      threshold: 0.80
      min_area: 1500
      max_area: 100000

audio:
  enabled: true
  max_not_heard: 15
  min_volume: 500
  listen:
    - bark
    - fire_alarm
    - scream
    - speech
    - yell
    - car_alarm
    - siren
    - alarm_clock
  filters:
    speech:
      threshold: 0.8

record:
  enabled: true
  motion:
    days: 10

  alerts:
    retain:
      days: 10
      mode: motion
  detections:
    retain:
      days: 10
      mode: motion

snapshots:
  enabled: true
  timestamp: true
  bounding_box: true
  retain:
    default: 10

cameras:
  entrada:  
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/entrada_sub  # H.264 substream
          hwaccel_args: preset-jetson-h264
          input_args: preset-rtsp-restream-low-latency
          roles: [detect, audio]
        - path: rtsp://127.0.0.1:8554/entrada
          hwaccel_args: preset-jetson-h264
          input_args: preset-rtsp-restream-low-latency
          roles: [record]
    detect:
      width: 640
      height: 360
      fps: 5
    zones:
      entradaz0:
        coordinates: 
          0.002,0.965,0.303,0.906,0.303,0.431,0.899,0.399,0.943,0.84,0.999,0.822,0.999,0.999,0.001,0.999
        loitering_time: 0
        objects:
          - person
          - dog
          - cat
          - bicycle
        friendly_name: EntradaZ0
    review:
      alerts:
        labels: [person, car, dog, cat]
        required_zones: entradaz0
      detections: {}
  

semantic_search:
  enabled: true
  model_size: small

face_recognition:
  enabled: true
  model_size: small

lpr:
  enabled: false

camera_groups:
  Cameras:
    order: 1
    icon: LuEye
    cameras:
      - birdseye
      - entrada
      - entrada2
      - entrada4

docker-compose file or Docker CLI command

services:
  mqtt:
    container_name: mqtt
    image: eclipse-mosquitto:2
    restart: unless-stopped
    ports:
      - "1883:1883"
    volumes:
      - /opt/frigate/mqtt/config:/mosquitto/config
    networks:
      - frigate_net

  frigate:
    container_name: frigate
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable-tensorrt-jp6
    runtime: nvidia
    privileged: true
    shm_size: "1gb"
    depends_on:
      - mqtt                    
    environment:
      - FRIGATE_RTSP_PASSWORD=XXXX
      - TZ=America/Recife
      - YOLO_MODELS=yolov7-320
      - USE_FP16=true
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,video,utility
    devices:
      - /dev/dri:/dev/dri
    ports:
      - "8971:8971"
      - "8554:8554"
      - "8555:8555/tcp"
      - "8555:8555/udp"
      - "5000:5000"
    volumes:
      - /opt/frigate/config:/config
      - /opt/frigate/media:/media/frigate
      - /opt/frigate/labelmap:/labelmap
      - /opt/frigate/config/model_cache:/config/model_cache
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    networks:
      - frigate_net

Relevant Frigate log output

info | 2026-03-29 21:43:15 | startup                            | Preparing Frigate...
info | 2026-03-29 21:43:16 | startup                            | Starting Frigate...
info | 2026-03-29 18:43:23 | frigate.util.config                | Checking if frigate config needs migration...
info | 2026-03-29 18:43:23 | frigate.util.config                | frigate config does not need migration...
info | 2026-03-29 18:43:23 | frigate.app                        | Starting Frigate (0.17.0-f0d69f7)
info | 2026-03-29 18:43:23 | peewee_migrate.logs                | Starting migrations
info | 2026-03-29 18:43:23 | peewee_migrate.logs                | There is nothing to migrate
info | 2026-03-29 18:43:23 | frigate.app                        | Recording process started: 764
info | 2026-03-29 18:43:23 | frigate.app                        | Review process started: 777
info | 2026-03-29 18:43:23 | frigate.app                        | go2rtc process pid: 177
info | 2026-03-29 18:43:24 | frigate.app                        | Embedding process started: 787
info | 2026-03-29 18:43:24 | frigate.app                        | Output process started: 838
info | 2026-03-29 18:43:24 | frigate.detectors.plugins.tensorrt | Loaded engine size: 73 MiB
info | 2026-03-29 18:43:24 | frigate.audio_manager              | Audio processor started (pid: 872)
info | 2026-03-29 18:43:24 | frigate.camera.maintainer          | Camera processor started for entrada: 876
info | 2026-03-29 18:43:24 | frigate.detectors.plugins.tensorrt | [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +5, GPU +8, now: CPU 182, GPU 2811 (MiB)
info | 2026-03-29 18:43:24 | frigate.detectors.plugins.tensorrt | [MemUsageChange] Init cuDNN: CPU +1, GPU +3, now: CPU 183, GPU 2814 (MiB)
info | 2026-03-29 18:43:24 | frigate.detectors.plugins.tensorrt | [MemUsageChange] TensorRT-managed allocation in engine deserialization: CPU +0, GPU +70, now: CPU 0, GPU 70 (MiB)
info | 2026-03-29 18:43:24 | frigate.detectors.plugins.tensorrt | [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +0, GPU +1, now: CPU 109, GPU 2739 (MiB)
info | 2026-03-29 18:43:24 | frigate.detectors.plugins.tensorrt | [MemUsageChange] Init cuDNN: CPU +1, GPU +0, now: CPU 110, GPU 2739 (MiB)
info | 2026-03-29 18:43:24 | frigate.detectors.plugins.tensorrt | [MemUsageChange] TensorRT-managed allocation in IExecutionContext creation: CPU +0, GPU +14, now: CPU 0, GPU 84 (MiB)
info | 2026-03-29 18:43:24 | frigate.camera.maintainer          | Capture process started for entrada: 960
info | 2026-03-29 18:43:24 | frigate.camera.maintainer          | Camera processor started for entrada2: 1029
info | 2026-03-29 18:43:24 | frigate.api.fastapi_app            | Starting FastAPI app
info | 2026-03-29 18:43:25 | frigate.api.fastapi_app            | FastAPI started
info | 2026-03-29 18:43:25 | frigate.camera.maintainer          | Capture process started for entrada2: 1115
info | 2026-03-29 18:43:25 | frigate.camera.maintainer          | Camera processor started for entrada4: 1192
info | 2026-03-29 18:43:25 | frigate.camera.maintainer          | Capture process started for entrada4: 1249
info | 2026-03-29 18:55:04 | frigate.comms.dispatcher           | Turning on detection for entrada

Install method

Docker Compose

Object Detector

TensorRT

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

Your configuration looks mostly correct for a Jetson Orin Nano with the TensorRT detector. The logs show the TensorRT engine loaded successfully and detection was turned on for the entrada camera. (1)

There are a few issues to address:

1. Detection is not enabled by default in your config

You need to explicitly enable detection in your camera configuration. Add enabled: true under the detect section for each camera: (1)

cameras:
  entrada:
    detect:
      enabled: true
      width: 640
      height: 360
      fps: 5

(1)

2. Jetson is community supported

Jetson devices are community supported, meaning support may be limited. (1), (2)

3. Your TensorRT configuration looks correct

Your detector and model configuration matches the documented format: (1)

detectors:
  tensorrt:
    type: tensorrt
    device: 0

model:
  path: /config/model_cache/tensorrt/yolov7-320.trt
  labelmap_path: /labelmap/coco-80.txt
  input_tensor: nchw
  input_pixel_format: rgb
  width: 320
  height: 320

(1)

4. Review configuration issue

Your review.detections is empty ({}), which may prevent detections from being retained. Add labels you want to track: (1)

review:
  alerts:
    labels: [person, car, dog, cat]
    required_zones: entradaz0
  detections:
    labels: [person, car, dog, cat, bird, bicycle]

The log entry Turning on detection for entrada suggests detection was manually enabled after startup, confirming it wasn't enabled by default in your config.

📚 Sources:

1. Supported Hardware
2. can frigate give a docker for origin nano+jetpack6.2.1? we need run cuda+onnx on this platform

---

Have another question? Just tag @inkeep.

[hawkeye217]

The AI is correct. Just add enabled: true to the detect section of your config.