Brenholder
Members
-
Joined
-
Last visited
-
OK, I've made some progress. detectors: # coral: # type: edgetpu # device: pci:0 tensorrt: type: tensorrt device: 0 #This is the default, select the first GPU model: path: /trt-models/yolov4-tiny-416.trt input_tensor: nchw input_pixel_format: rgb width: 416 height: 416 I had the config file like this. The detector doesn't work, but there are no errors in the log. Uncommenting the coral lines and the coral detector with start working. detectors: # coral: # type: edgetpu # device: pci:0 tensorrt: type: tensorrt device: 0 #This is the default, select the first GPU model: path: /trt-models/yolov4-tiny-416.trt input_tensor: nchw input_pixel_format: rgb width: 416 height: 416 This config does work, the tensorrt detector events start showing up. But the there is an issue when trying to use the coral detector. Uncommenting those lines will cause the coral to not be detected At this point I can have the coral tpu or the GPU detector running, but not both.
-
I turned person down to 10% and still nothing. I went back to the logs from the docker that generates the .trt files there were a bunch of lines Does that mean I should disable FP16?
-
No detections. Watching the debug view I can see the bounding box for motion but no detections. The GPU shows up on the system page as a detector. Using the same config file if enable a TPU detector the detection works and I can see the bonding boxes in the debug view. So I think I have the camera and object sections setup correctly.
-
The models are in the path I don't see anything in the log that tripped me off
-
I have my RTX 3070 setup in frigate to transcode and do tensor detection. Transcoding it working and the card shows up as a detector but it isn't triggering any events. Events do trigger correctly using a Coral TPU. Is there something else I need to configure to get tensorrt events working?
-
Is there way to adjust the delay between episodes when they are auto playing? 15 seconds seems long.
-
Just wanted to say that your logs and FAQ are great. Makes it so easy to fix issues that come up. Thanks.