Hi
I m running Doods on Hassio with the Google Coral Stick.
I use this model: ssd_mobilenet_v2_coco_quant_postprocess_edgetpu.tflite
the system is running great but i get this message in the logs:
Updating doods image_processing took longer than the scheduled update interval
the scan interval is 0.5 seconds (doods logs shows “duration”: 0.060507047) i have 5 cameras connected.
how can i improve the speed?
I keep getting your error, as well as this:
2021-03-16 15:49:49 ERROR (MainThread) [homeassistant.components.image_processing] Error on receive image from entity: Unable to get image
Is there an issue with doods currently?
I am trying to use an mjpeg stream from an ESP32-CAM, and the stream shows up fine in HA with good FPS.
actually i don’t know whats the problem is.
no i set the scan_interval to 2 seconds. but sometime it takes more than 5 seconds to complete the next picture scan
i would like to achieve a speed of detection like in this video
I recently change all my cameras and notice that processing times seem to increase and I missing many detections so I begin checking
I ended up setting “environment variable” for docker container logger_level = debug
with this I find below
2021-04-27T21:02:10.381-0500 DEBUG tflite/detector.go:256 Decoded Image {"package": "detector.tflite", "name": "edgetpu", "id": "", "width": 2592, "height": 1944, "duration": 0.063528916}
2021-04-27T21:02:10.382-0500 DEBUG tflite/detector.go:259 Resized Image {"package": "detector.tflite", "name": "edgetpu", "id": "", "width": 300, "height": 300, "duration": 0.064048015}
2021-04-27T21:02:10.382-0500 DEBUG tflite/detector.go:274 Image pre-processing complete {"package": "detector.tflite", "name": "edgetpu", "duration": 0.064343871}
2021-04-27T21:02:10.389-0500 DEBUG tflite/detector.go:321 Inference complete {"package": "detector.tflite", "name": "edgetpu", "inference_time": 0.007067734, "duration": 0.071497608}
2021-04-27T21:02:10.389-0500 INFO tflite/detector.go:399 Detection Complete {"package": "detector.tflite", "name": "edgetpu", "id": "", "duration": 0.0715886, "detections": 20, "device": {"Type":0,"Path":"/dev/apex_0"}}
debug show more about what happening with image
decode approx up to 0.06 seconds
resize up to 0.06 seconds
preprocess up to 0.06 seconds
inference up to .007
detection complete up to 0.06 seconds
so I expecting total process times of 0.1 second total at most but look more like time is up to 0.3 seconds 0r more
One thing I remember is that doods is expecting certain image size 300x300 and must resize image to fit this. My camera offer (2) stream, high def and low so I plan to resize low def size to 300x300 and use that as input to doods and see if this reduce process times.
Also, disk writes could be issue as well since it is necessary to store image after complete. In this case you may need to consider how this affect the time.