AV1 Encoding Support?

[moogoos]

I currnelty have a bunch of hikvision cameras and have to record in h264 so I dont have to transcode saved files for iOS and can view the feeds in a browser. I'd like to explore transcoding the h264 streams into AV1 to save space and maintain current capability if possible. Is this possible? I have an ARC A310 card, an NVIDIA 5060Ti to mess around with and also will be toying around with an Arc B50. I have 8 cameras.

[inkeep-ai-assistant[bot]]

The documentation I have access to focuses on object detection hardware support rather than video encoding/transcoding capabilities(1). While it confirms that Intel Arc GPUs and NVIDIA GPUs are supported for object detection through OpenVINO and TensorRT respectively(1), it doesn't provide information about AV1 encoding support or transcoding configuration.

For AV1 encoding and transcoding configuration with your Intel Arc A310, NVIDIA RTX 4060Ti, or Intel Arc B580, you'll need to check Frigate's video processing and recording configuration documentation, which isn't covered in the available sources.

I'd recommend visiting GitHub for more specific information about video encoding capabilities and transcoding configuration.

Would you like a general answer based on my knowledge instead?

📚 Sources:

1. Supported Hardware

---

Have another question? Just tag @inkeep.

[NickM-27]

There is no requirement to use h264 for iOS. I use h265 on all my cameras without any issues

[moogoos]

I cant download and play files in that format

[NickM-27]

Have you enabled Apple compatibility? https://docs.frigate.video/configuration/camera_specific#h265-cameras-via-safari

[moogoos]

oh i use chrome. never use safari. I'll take a look.

What about AV1 encoding? Is that possible on the output stream to record maintaining quality?

[NickM-27]

It's possible, but not really a great option as it would run continuously and use GPU resources.

[moogoos]

I have a relatively low power GPU 70w so im willing to try it out at least. Any guides for that somewhere?

[pavelkms]

Settings of one of my cameras for direct encoding to av1:

Podezd:
  genai:
    enabled: true
  birdseye:
    order: 1
  face_recognition:
    enabled: true
  ffmpeg:
    hwaccel_args: preset-intel-qsv-h264
    output_args:
      record: -f segment -segment_time 10 -segment_format mp4 -reset_timestamps 1 -strftime 1 -c:v av1_qsv -global_quality:v 31 -c:a aac -b:a 64k
    inputs:
      - path: rtsp://192.168.0.111:8554/Podezd
        input_args: preset-rtsp-restream-low-latency
        roles:
          - record
          - detect
  notifications:
    enabled: true
  motion:
    mask: 0,0.048,0.24,0.046,0.242,0.1,0,0.102
    threshold: 70
    contour_area: 10
    improve_contrast: true
  live:
    streams:
      Main Stream: Podezd
      Sub Stream: Podezd_sub

[moogoos]

how well does this work? Quality the same? Any idea what kind of storage savings you're getting?

[pavelkms]

This works great on the Arc a310. The quality is the same. Space savings largely depend on your camera's stream. I have five different cameras with different streams. The savings range from 5x to 2x. You can always try it out and see for yourself.

[moogoos]

thanks. yeah that's the plan just trying to gauge benefit first...