robmarkcole / coral-pi-rest-server Goto Github PK

Hi robmarcole,
I tried to use your docker, but when I start it I receive the following errors:

Traceback (most recent call last):
File "coral-app.py", line 84, in
engine = DetectionEngine(MODEL_FILE)
File "/usr/local/lib/python3.6/site-packages/edgetpu/detection/engine.py", line 72, in init
super().init(model_path)
File "/usr/local/lib/python3.6/site-packages/edgetpu/basic/basic_engine.py", line 40, in init
self._engine = BasicEnginePythonWrapper.CreateFromFile(model_path)
RuntimeError: Could not open '/home/robin/edgetpu/all_models/mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'.

please could you help me to understand how can I fix this issue on my debian?
Thank you

https://coral.withgoogle.com/docs/edgetpu/retrain-classification-ondevice/

Classification is suitable in the home automation context where a typical use case is to filter images captured by motion activated cameras, or determine some binary state (e.g. garrage door open/closed). The drawbacks of this technique are not a problem in this context, where we typically dont want to have to create massive datasets, and also we expect datasets to have a small inner class variation. The docs state 'However, this method works well when the training images (and the images to classify) have less variations. And in such case, only a few images (less than 10) are needed to train. For example, when an assembly-line produces a new type of defective part, you can use this method to teach the device what a defective part looks like on the fly.'

is it also possible to use this server with an amd64 system?

I have tried to build the docker container of this fork:
https://github.com/lmamakos/coral-pi-rest-server

But I get this error Your platform is not supported.

I think that requirements.txt needs to include numpy. While starting coral-app.py inside a container with only Flask and Pillow installed I note that something in the edgetpu library has a dependency on it:

root@8d34eb5617d9:/home/pi# python coral-app.py
ImportError: numpy.core.multiarray failed to import
Traceback (most recent call last):
  File "coral-app.py", line 6, in <module>
    from edgetpu.detection.engine import DetectionEngine
  File "/usr/local/lib/python3.6/site-packages/edgetpu/detection/engine.py", line 17, in <module>
    from edgetpu.basic.basic_engine import BasicEngine
  File "/usr/local/lib/python3.6/site-packages/edgetpu/basic/basic_engine.py", line 17, in <module>
    from edgetpu.swig.edgetpu_cpp_wrapper import BasicEnginePythonWrapper
  File "/usr/local/lib/python3.6/site-packages/edgetpu/swig/edgetpu_cpp_wrapper.py", line 28, in <module>
    _edgetpu_cpp_wrapper = swig_import_helper()
  File "/usr/local/lib/python3.6/site-packages/edgetpu/swig/edgetpu_cpp_wrapper.py", line 24, in swig_import_helper
    _mod = imp.load_module('_edgetpu_cpp_wrapper', fp, pathname, description)
  File "/usr/local/lib/python3.6/imp.py", line 243, in load_module
    return load_dynamic(name, filename, file)
  File "/usr/local/lib/python3.6/imp.py", line 343, in load_dynamic
    return _load(spec)
ImportError: numpy.core.multiarray failed to import

Adding that back to requirements.txt resolves this problem.

Raspbian has python 3.5.3 but this is old

Hit issue f0cal/google-coral#42 but did not resolve by removing refs to edgetpu. Working with this hardware is frustrating

Promise to be more performant https://coral.withgoogle.com/models

See -> https://github.com/dgomes/cusca/blob/master/app.py

Partner with @dgomes

As title

As title

I've noticed that in my environment, the coral TPU stick goes into an unusual state and the flask server ceases responding to requests. I'm running this in a Docker container on Ubuntu 18.04 on Intel Core i5.

Of note that when it gets into this state, the white LED on the Coral TPU USB stick has this consistent blinking going on. I've tried to figure out what the LED state is supposed to represent, but haven't obviously found documentation to that effect. Any insight here would be useful.

I've added a healthcheck to the docker container, which successfully noticed that the server isn't responding to queries. Too bad it doesn't automatically restart the container :-) I think that I need to be running docker swarm or something a bit more sophisticated for the automatic restart...

Of course I'm away on travel, so debugging is limited. I need to sync-up versions of what I'm running with the most recent version here, and also look at some suspicious/weirdness in terms of what packages are being installed into the container by the coral install script. I fear there might be a mix of Python 3.6 and 3.7 stuff going on in there, and I'm not 100% clear on what's being used. Of course, I'd think that obvious dependency issues would be evident right away, not days later after the container has been running.

I wanted to just open an issue here as a place-keeper and also see if anyone else might have noticed similar behavior in an attempt to shorten up some debugging steps.

Add a folder with demo scripts showing the usage of the rest-server. Ideas include

• Script which takes an image, processes it, and generates an annotated jpg

• Script for exposing the image as mjpeg stream/posting over mqtt

As title, appears the coral build scripts haven't been updated for Buster yet. Track progress here. My error:

pi@tpu-pi:/usr/local/lib/python3.7/dist-packages/edgetpu/demo $ python3 classify_image.py \
> --model ~/Downloads/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite \
> --label ~/Downloads/inat_bird_labels.txt \
> --image ~/Downloads/parrot.jpg
Traceback (most recent call last):
  File "/usr/local/lib/python3.7/dist-packages/edgetpu/swig/edgetpu_cpp_wrapper.py", line 18, in swig_import_helper
    fp, pathname, description = imp.find_module('_edgetpu_cpp_wrapper', [dirname(__file__)])
  File "/usr/lib/python3.7/imp.py", line 296, in find_module
    raise ImportError(_ERR_MSG.format(name), name=name)
ImportError: No module named '_edgetpu_cpp_wrapper'

As title e.g. 0.9979328

Display last processed and metadata on a web page, use flask

Display stream with https://github.com/robmarkcole/simple_mjpeg_streamer_http_server

As title, so that I don't have to maintain a Home Assistant support library https://deepstackpython.readthedocs.io/en/latest/objectdetection.html

I tried posting a completely blank (black) image to see if this would crash the server, but it returned the predictions from a previous image (containing people).

https://github.com/gstaff/flask-ngrok

I am getting:

(base) robin@robins-ubuntu:~/Github/coral-pi-rest-server$ sudo docker run -p 5000:5000 --device /dev/bus/usb:/dev/bus/usb   coral:latest
Traceback (most recent call last):
  File "coral-app.py", line 122, in <module>
    load_model(args.model, args.labels)
  File "coral-app.py", line 48, in load_model
    engine = DetectionEngine(model_file)
  File "/usr/local/lib/python3.6/site-packages/edgetpu/detection/engine.py", line 55, in __init__
    super().__init__(model_path)
  File "/usr/local/lib/python3.6/site-packages/edgetpu/swig/edgetpu_cpp_wrapper.py", line 300, in __init__
    this = _edgetpu_cpp_wrapper.new_BasicEngine(*args)
RuntimeError: Failed to allocate tensors.

Any ideas @lmamakos ?

The package below does this, might be a nice way to integrate with home assistant and expose results of processing

https://github.com/blakeblackshear/frigate

Can install from https://www.coral.ai/software/

add to requirements or setup.py

GOAL: subscribe to rtsp/mjpeg stream and display processed frame, suitable for low FPS with no caching of data.
OPTIONAL: for batch processing, put frames on a REDIS queue using separate process

• Complete code from this tutorial
• Capture frames and place on Redis -> https://github.com/robmarkcole/hello-websocket/tree/fix-python3
• Frigate also has example of capturing from stream with openCV

Display stream issue -> #14

Server times out and I see

192.168.1.164 - - [10/Aug/2019 18:37:31] "POST /v1/vision/detection HTTP/1.1" 200 -
<FileStorage: 'image' (None)>
192.168.1.164 - - [10/Aug/2019 18:39:03] "POST /v1/vision/detection HTTP/1.1" 200 -
<FileStorage: 'image' (None)>
192.168.1.164 - - [11/Aug/2019 09:56:44] "POST /v1/vision/detection HTTP/1.1" 200 -
<FileStorage: 'image' (None)>
<FileStorage: 'image' (None)>

as title

https://github.com/Kao1126/EdgeTPU-FaceNet

As title

• https://www.tensorflow.org/tfx/serving/api_rest
• https://cloud.google.com/community/tutorials/standalone-tensorflow-raspberry-pi

Hi - Im getting the error message that you noted in your troubleshooting. It says to reflash the SD card. What does reflash the SC card mean?

Q: I get the error: HandleQueuedBulkIn transfer in failed. Not found: USB transfer error 5 [LibUsbDataInCallback] A: I reflashed the SD card and tried again with success

Thanks.

• Tests

• Basic username/password auth

• Server status endpoint

As per #13 you cant currently use the coral with system-wide python in Buster, so the recommended solution is Docker on the pi. @lmamakos can we have a single docker image that will work on Ubuntu & Raspbian? If not will we require hardware specific instructions on building the container?

https://blogs.sap.com/2020/02/11/containerizing-a-tensorflow-lite-edge-tpu-ml-application-with-hardware-access-on-raspbian/

Add web UI for uploading images -> https://github.com/render-examples/fastai-v3/blob/master/app/server.py

Add service to run in background

For some bizarre reason commits are labeled as from @mtyka

re: Coral rest server

there any way to make the font larger and/or a different color background on the boxes? I’m not able to read anything on any detections, possibly b/c I’ve reduced the camera image to VGA resolution (and the bitrate is now 96kbps on the sub stream) as its a wifi camera and the link isn’t great. This solved some of my delay problems and timeouts:

I’d like to move them to the bottom instead of the top of the image as well.

If you can just point me to the module that handles these drawing routines, I should be able to figure it out from there.

Jeff

https://github.com/photoprism/photoprism

For a 'front end'?

To allow processing of a Redis queue

As title. Expose logs on an endpoint

Use if flask.request.method != "POST" or not flask.request.files.get("image"):, do the negative. This saves indentation. Same for if not predictions

https://refactoring.com/catalog/replaceNestedConditionalWithGuardClauses.html

As title, to expand platforms supported and remove requirement for coral hardware

This seems to be the most popular request, but requires some preprocessing of the images

References

• https://thedatamage.com/face-recognition-tensorflow-tutorial/

• https://machinelearningmastery.com/how-to-develop-a-face-recognition-system-using-facenet-in-keras-and-an-svm-classifier/

• https://coral.withgoogle.com/examples/detect-image/#perform-face-detection

As title

We had /predict, now /v1/vision/detection. Makes sense to have this as configurable as this would allow people to easily integrate the endpoints into their existing flask apps

As title, to allow object & face

following your guide, i'm trying to run python3 coral-app.py but i get this:

pi@raspberrypi:~/coral-pi-rest-server $ python3 coral-app.py ImportError: numpy.core.multiarray failed to import Traceback (most recent call last): File "coral-app.py", line 6, in <module> from edgetpu.detection.engine import DetectionEngine File "/usr/local/lib/python3.5/dist-packages/edgetpu/detection/engine.py", line 17, in <module> from edgetpu.basic.basic_engine import BasicEngine File "/usr/local/lib/python3.5/dist-packages/edgetpu/basic/basic_engine.py", line 17, in <module> from edgetpu.swig.edgetpu_cpp_wrapper import BasicEnginePythonWrapper File "/usr/local/lib/python3.5/dist-packages/edgetpu/swig/edgetpu_cpp_wrapper.py", line 28, in <module> _edgetpu_cpp_wrapper = swig_import_helper() File "/usr/local/lib/python3.5/dist-packages/edgetpu/swig/edgetpu_cpp_wrapper.py", line 24, in swig_import_helper _mod = imp.load_module('_edgetpu_cpp_wrapper', fp, pathname, description) File "/usr/lib/python3.5/imp.py", line 242, in load_module return load_dynamic(name, filename, file) File "/usr/lib/python3.5/imp.py", line 342, in load_dynamic return _load(spec) ImportError: numpy.core.multiarray failed to import

i'm on standard Raspbian GNU/Linux 9 (stretch) ( Raspberry 3+)
i've installed the Edge TPU runtime and Python library from the coral site
any idea?

this is my home structure:

pi@raspberrypi:~ $ ls
all_models            env               start_assistant.sh
coral-pi-rest-server  googleassistant   Templates
Desktop               Music             tv_grab_italy_basic_url.py
Documents             Pictures          tv_grab_italy_sky_url.py
Downloads             Public            Videos
edgetpu_api           Python-3.6.8
edgetpu_api.tar.gz    Python-3.6.8.tgz

With the help of https://github.com/guichristmann/edge-tpu-tiny-yolo
I made a tiny-yolo version of the rest-server:
https://gist.github.com/weltmeyer/22ede027161ec13baf0671151399e726

if you are interested in this.

I made this change:

--- coral-app.py.original	2019-06-07 00:07:28.942209359 -0400
+++ coral-app.py	2019-06-07 00:22:49.023841467 -0400
@@ -83,8 +83,8 @@

                     bounding_box = {
                         "x1": round(prediction.bounding_box[0, 0], DECIMALS),
-                        "x2": round(prediction.bounding_box[0, 1], DECIMALS),
-                        "y1": round(prediction.bounding_box[1, 0], DECIMALS),
+                        "x2": round(prediction.bounding_box[1, 0], DECIMALS),
+                        "y1": round(prediction.bounding_box[0, 1], DECIMALS),
                         "y2": round(prediction.bounding_box[1, 1], DECIMALS),
                     }
                     preds.append(

and now the bounding boxes seem to line up better..

Using sudo systemctl start coral.service the app appears to die after 12 hours, no errors in logs, just doesn't respond to requests

Can we support coral on mac via docker? Seems we should be able to, I'm not sure how we mount the USB however