Counting the dishes using detection and tracking algorithm

The objective is to count the plates using object detection and object tracking. The repository used the PyimageSearch Simple Object Tracking code frome here. The object detection model used is pre-trained Faster R-CNN model trained on Detectron2 framework and a custom dataset.

Output of the code counting the dishes!

Installation of the environment

We encourage you to use conda environment. Once you create an environment use follwoing command to get the environment ready

pip install -r requirements.txt

Training the Faster R-CNN Model

Follow instruction given in PlateCount_FasterRCNN.ipynb file for training the model. You can use the data from the roboflow account provided in the file.

Inference on Images and Videos

Run the command to test on an Image

python test_images2.py

Run the command to test on a Video

python test_video.py

Make sure to provide right path of video, model file and image file in the script.

Common Debug

1. Make sure the environment is properly installed
2. Provide complete path of the model file
3. Detectron2 installation will take time. Be patient.