blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:21 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@B0ndo2 commented on GitHub (Jan 29, 2025):

path: /frigate/efficientdet_lite3x_640_ptq_edgetpu.tflite

@lahayeg where do I download the file above ?

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:19 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@NickM-27 commented on GitHub (Sep 15, 2023):

Oh I see, yeah so it will be something that Nate will likely need to look into, I am not familiar with the script itself and if it supports selecting…

blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:19 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@lahayeg commented on GitHub (Sep 19, 2023):

Tested with efficientdet_lite3_512_ptq_edgetpu.tflite and same issue. I don't want to waste more time for everyone so I prefer to close it (won't do ;-)

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:19 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@vajonam commented on GitHub (Sep 15, 2023):

If we can update those manual commands to point to newer tensorrt repos, I could build that offline and provide it to the container which would not…

blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:19 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@lahayeg commented on GitHub (Sep 19, 2023):

lol : I restarted my RPI4 with

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:19 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@NateMeyer commented on GitHub (Sep 15, 2023):

Hey, that's a great point, I hadn't considered that issue before. For now, running once with just the detection GPU to build the models should be a…

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:16 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@vajonam commented on GitHub (Sep 15, 2023):

```docker run --gpus=all --rm -it -v pwd/trt-models:/tensorrt_models -v pwd/tensorrt_models.sh:/tensorrt_models.sh nvcr.io/nvidia/tensorrt:22.07-py…

blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:16 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@lahayeg commented on GitHub (Sep 19, 2023):

Sorry just saw the comment on Coral:

_* EfficientDet-Lite3x is not compatible with Edge TPUs over USB; it can be used only with PCIe-based devices.…

blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:13 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@lahayeg commented on GitHub (Sep 19, 2023):

Thanks @NickM-27 to pointing it

several tried: with the correct size (300 as defined by the Coral doc instead of the default 320)

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:13 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@NickM-27 commented on GitHub (Sep 15, 2023):

how exactly did you specify a GPU for the building of the model before?

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:11 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@vajonam commented on GitHub (Sep 15, 2023):

okay I thought I just made it up.. cool thanks! There are some deprecation warnings while building the model, safe to ignore?

blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:11 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@NickM-27 commented on GitHub (Sep 15, 2023):

If I set again the custom model, any risk to "burn" my coral?

I don't see why it would

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:11 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@vajonam commented on GitHub (Sep 15, 2023):

@NickM-27 I have multiple GPU's and I want to run the detection against GPU: 1, model is built on GPU 0, previously when I was manually building the…

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:11 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@vajonam commented on GitHub (Sep 15, 2023):

blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:11 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@lahayeg commented on GitHub (Sep 18, 2023):

tried again

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:11 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@NickM-27 commented on GitHub (Sep 15, 2023):

that would be a question for @NateMeyer

blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:11 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@lahayeg commented on GitHub (Sep 18, 2023):

tried to put model at /config/ level and still error in log

blu commented on issue starred/frigate-blakeblackshear-2#3854 2026-02-06 11:48:11 -08:00
[Config Support]: How to config and store Custom Model with Frigate Home Assistant add on

@NickM-27 commented on GitHub (Sep 18, 2023):

ValueError: Cannot set tensor: Dimension mismatch. Got 320 but expected 300 for dimension 1 of input 0.

looks like a config error now

blu commented on issue starred/frigate-blakeblackshear-2#3856 2026-02-06 11:48:11 -08:00
[TensorRT Support]: No way to select which GPU a model is built for when multiple GPUs are present

@vajonam commented on GitHub (Sep 15, 2023):

The reason is the one of them is a mining GPU, which is good for cuda tasks, and other GPU is used for encode/decode. Which also helps my other apps…