Raccoon is a TensorFlow based application designed to recognize people on video feeds. It uses the latest version of TensorFlow and is hosted on a custom Docker system consisting of multiple different Docker images. The object detector is trained on a custom dataset of approximately 1000 annotated images hosted at https://universe.roboflow.com/cuzimclicks/raccoon-qnjzn. The dataset contains three different classes and has been resized to 640x640.
- Docker
To use Raccoon, you will need to install TensorFlow. You can then clone the repository from GitHub using the following command:
git clone https://github.com/CuzImClicks/Raccoon.git
To use Raccoon, you will need to provide it with an input image or video stream. The object detector will then return the bounding boxes and labels for any detected objects.
You can run the object detection script using the following command:
python plot_object_detection_saved_model.py
For more options, you can use the following command to view the available arguments:
python examples/detect_image.py --help
To train the object detector, you will need to use the custom dataset of annotated images hosted at https://universe.roboflow.com/cuzimclicks/raccoon-qnjzn. This can be done using the TensorFlow training API. Make sure to specify the three different classes and that the images have been resized to 640x640.
Raccoon can be hosted using a custom Docker system consisting of multiple different Docker images, which are hosted at Docker Hub:
TensorFlow
The first version of this object detector, which is very accurate but slow on low-performance CPUs/GPUs
Edge TPU
the Edge TPU version of this object detector, which is quick but has low accuracy
Compiler
comes with the Edge TPU compiler preinstalled
Vanilla TensorFlow
Build the docker imagedocker build -t my-name/my-image .
Run the docker image
docker run --rm -i -t my-name/my-image bash
In the bash command line run to start the program
python plot_object_detection_saved_model.py
To copy something(the final images) to or from a docker container
docker ps
docker cp <file path> <container id>:<destination path>yaroslavvb/tensorflow-community-wheels#217
Run the same container again
docker container ps -a
Find the container you want
docker exec -it <CONTAINER_ID> bash
Input folder for copying with pull.sh
/share/Henrik/input
Output folder for copying with push.sh
/share/Henrik/output
Location of the converted YOLOV7 model
/share/Henrik/yolov7_model.tflite
Location of the security camera alarm images
/share/Public/record_nvr_images
tflite_convert --[keras_model_file|saved_model_dir]<input> --output_file=<file>
- train own dataset
- dockerize for deployment
- code quality with codacy
- tensorflow
- pr based contributing
- discord webhook for results
Links
'${\r' command not found -> apt install dos2unix