NOTE: This project won the first prize in AI Enterpreneural 2021, a deep learning competition organized by IIT Kharagpur and sponsored by Intel!
Contributors: Avirup Dey (Jadavpur University) and Sarosij Bose (University of Calcutta).
This repository contains two lightweight Traffic sign classification implementations which can predict Traffic signs from any real time video feed. Here, a model based on an slightly enhanced LeNet architecture has been used and trained on the German Traffic Sign Dataset (GTSD) which has over 70000 images of traffic signs and over 40 various classes. Our model achieves a validation accuracy of over 98% and a training accuracy of over 97%. This saved model is then optimized over the Intel OpenVINO Model Optimizer + Inference Engine and run directly for predicting Traffic signs live from any video source(we have used webcam for our run). We have also provided a non optimized solution for comparison purposes.
To learn more about the features which our model offers, it's advantages and how it can be socially useful in a number of applications, see Project Details
Please see the requirements.py file to know and install the necessary libraries.
Download and install the OpenVINO toolkit from Intel's website according to your operating system.
git clone https://github.com/sarobml2000/TSCLite.git
Run the notebook as given below. The created tsc_model.h5 file will be used later for prediction.
cd TSCLite\code
jupyter notebook TSC_Tensorflow.ipynb
The compiled model(in .h5 format) is avaliable within the dependencies folder.
cd TSCLite\dependencies
Copy the labels.csv file into the working directory from the dependencies folder. Run the notebook.
cd TSCLite\code
jupyter notebook TSC_OpenVINO.ipynb
After converting the model to ONNX format, run the following commands in windows terminal.
Initialize OpenVINO environment variables.
path to openvino\bin\setupvars.bat
For linux based systems, use this.
cd path to openvino/bin
source ./setupvars.sh
Generate the .xml and .bin files by running the Inference engine.
path to openvino\deployment_tools\model_optimizer\mo.py --input_model <path to model.onnx> --input_shape[1,32,32,1]
Once the .xml and .bin files are created, cd them to your working directory and run the rest of the code. Make sure you provide the correct
path to these files in the following lines of the notebook:
ie = IECore()
net = ie.read_network(model=r"<path to xml file>", weights=r"<path to bin file>")
exec_net = ie.load_network(network=net, device_name="CPU")
These files are also provided inside the external folder.
cd TSCLite\external
Copy the labels.csv file into the working directory from the dependencies folder. Run the notebook.
cd TSCLite\code
jupyter notebook TSC_without_OpenVINO.ipynb
Reduced the number of training epochs to achieve standard results. We repeatedly tweaked the model until attaining optimum results.
Here's the loss and accuracy plotted against the number of epochs.
The project was carried out with the following system specifications,
OS: Windows 10 64 bit Home.
CPU: Intel i5 8th Generation.
RAM: 8 GB
GPU: No hardware acceleration required for this project.
Mail us at [email protected] or [email protected]