Road_detection

Instructions to set up

1.Clone this repo

git clone https://github.com/csoham96/Road_detection.git

2.Install CUDA

3.Install pytorch

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

4.Install dependencies

pip install -r requirements.txt

5.Download Dataset from Kaggle (https://www.kaggle.com/datasets/eyantraiit/semantic-segmentation-datasets-of-indian-roads)

6.Training the model:

For training i have used the training set which consists around 2400 images,trained with th unet model for 30 epochs Achieved a Dice score of 80.97%. Specify the path of dataset inside train.py file

python train.py --amp -e 30

7.Converting into mobile-format run the file mobile_format.py which first converts model into torchscript and then into mobile format

python mobile_format.py

8.To infer on videos run infer_video.py

python infer_video.py