Does this setup still require a coral accelerator?

[faulkiebear]

The hardware requirements are not mentioned anywhere in this repo but inspecting the docker files this project still seems to depend on a coral accelerator. Or at the very least its mentioned alot in the docker files, frigate configs and docker run commands. Is this because I need one for it to work or because this relies on coral dependancies which also allow the nano gpu to work?

I've been trying to get this working on a jetson nano assuming that this might allow me to run the inference hardware acceleration on the jetson's internal tegra gpu.

Although I did manage to get the hardware acc'ed ffmpeg working (which takes an hour to build which I think should also be mentioned in the README, I also had to fiddle with the docker files a bit to make it successfully build), which seems to be prooved to work using the command: ffmpeg -c:v h264 -re -stream_loop -1 -i https://streams.videolan.org/ffmpeg/incoming/720p60.mp4 -f rawvideo -pix_fmt yuv420p pipe: > /dev/null Which has a much lower cpu load (when running top or jtop) than running without the first argument: ffmpeg -re -stream_loop -1 -i https://streams.videolan.org/ffmpeg/incoming/720p60.mp4 -f rawvideo -pix_fmt yuv420p pipe: > /dev/null

But the jetson_frigate build just wouldnt successfully complete, always exiting with: Step 22/27 : COPY --from=web /opt/frigate/build web/ COPY failed: stat opt/frigate/build: file does not exist Makefile:42: recipe for target 'jetson_frigate' failed

I've tried messing with the paths there in the "docker/Dockerfile.base" file as the 'COPY failed' line seems to be missing the root '/' before opt. But that didnt work. And there doesnt seem to be a way to list the directories inside the "web" container (as it's never been started) to see if the folders even exist.

If anyone has any ideas I'd be greatful. And do tell me if this has been pointless for me as it all depends on having a coral dongle.

My goal is to have 4 rtsp streams with inference and NVR capabilities on the jetson-nano. Frigate seems like the best option for this but maybe its not possible without a coral TPU?

[Eyal51]

I had the same issue. Any solution?

[ACiDGRiM]

Frigate docs indicate that it does not support GPU for object detection, just video decoding. Adding jetson for AI would need to be discussed on the upstream project. I believe that it's mentioned that access to AI compute is impeded by nVidia, which is likely true for the desktop cards.