[Config Support]: Config Validation Error with ONNX Detector on Nvidia/amd64

[Rishabh-Bajpai]

Describe the problem you are having

Hello,

I'm running into a persistent configuration validation error when trying to enable an ONNX detector for my Nvidia GPU and I'm hoping someone can spot what I'm missing.

The Problem:
Whenever I add the detectors section to my config.yml to use the ONNX detector, Frigate fails to start with the following validation error:

Could not determine exact line number: 'onnx'
Message : Input should be a valid string

What I've Tried:
Here's a summary of my troubleshooting:

1. CPU-only works: A minimal configuration with no GPU settings works perfectly.
2. GPU HWACCEL works: A configuration with ffmpeg: hwaccel_args: preset-nvidia-h264 and the correct docker-compose GPU settings also works, but it correctly falls back to using the CPU for detection. This confirms the container has access to the GPU.
3. Detector fails: The validation error Could not determine exact line number: 'onnx' only appears when I add the detectors: block to the config file.

I have used multiple methods (cat, echo, nano) to create the config.yml file to be certain there are no hidden tab characters or syntax errors. The problem seems to be isolated entirely to the detectors: section.

I'm at a dead end. It feels like the config is correct according to the documentation, but the validator disagrees. Could this be a bug, or am I missing something obvious?

Thank you for any help you can provide.

Version

0.16.1-e664cb2

Frigate config file

mqtt:
  enabled: False

detectors:
  nvidia:
    type: onnx
    device: 0

ffmpeg:
  hwaccel_args: preset-nvidia-h264

cameras:
  camera_1:
    ffmpeg:
      inputs:
        - path: rtsp://...
          roles:
            - detect
            - record
    detect:
      width: 640
      height: 480
    record:
      enabled: True
      retain:
        days: 7

  camera_2:
    ffmpeg:
      inputs:
        - path: rtsp://...
          roles:
            - detect
            - record
    detect:
      width: 640
      height: 480
    record:
      enabled: True
      retain:
        days: 7

Relevant Frigate log output

*************************************************************
*** Your config file is not valid!                     ***
*** Please check the docs at                           ***
*** https://docs.frigate.video/configuration/          ***
*************************************************************
*************************************************************
*** Config Validation Errors                           ***
*************************************************************

Could not determine exact line number: 'onnx'
Message : Input should be a valid string

*************************************************************
*** End Config Validation Errors                       ***
*************************************************************
[INFO] Service Frigate exited with code 1 (by signal 0)

Relevant go2rtc log output

NA

Frigate stats

No response

Operating system

Debian

Install method

Docker Compose

docker-compose file or Docker CLI command

services:
  frigate:
    container_name: frigate
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable
    shm_size: 1g
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - ./config:/config
      - ./storage:/media/frigate
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "5011:5000"

Object Detector

CPU (no coral)

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

The issue is with your Docker image and device configuration. For NVIDIA GPU support with ONNX, you need to use the -tensorrt Frigate image (1). Your current configuration uses the standard stable image which doesn't include the necessary TensorRT libraries for GPU acceleration.

Docker Image Fix:
Change your docker-compose to use the TensorRT image:

image: ghcr.io/blakeblackshear/frigate:stable-tensorrt

(1)

Device Configuration:
The device parameter should be a string, not an integer (1). Update your detector configuration:

detectors:
  nvidia:
    type: onnx
    device: '0'

(1)

Missing Model Configuration:
You're also missing the model configuration section. For ONNX detectors, you need to specify the model details. Here's an example for YOLO-NAS (2):

model:
  model_type: yolonas
  width: 320
  height: 320
  input_pixel_format: bgr
  input_tensor: nchw
  path: /config/yolo_nas_s.onnx
  labelmap_path: /labelmap/coco-80.txt

(2)

Why This Happens:
The validation error occurs because the standard Frigate image doesn't recognize ONNX as a valid detector type for GPU acceleration. TensorRT will automatically be detected and used as a detector in the -tensorrt Frigate image when a supported ONNX model is configured (3).

The -tensorrt image includes the necessary libraries for both TensorRT and ONNX GPU acceleration on NVIDIA hardware (1).

📚 Sources:

1. Supported Hardware
2. Supported Hardware
3. Supported Hardware

---

Have another question? Just tag @inkeep.

[Rishabh-Bajpai]

I was missing the model file as well.