This project is a web application that predicts the drag coefficient (Cd) of a vehicle based on its 3D point cloud data. The application is built with Streamlit and served using Docker.
The models and methods used in this project are based on the following research paper. The paper provides a detailed explanation of the data processing, model architecture, and experimental results.
→ Read the Full Research Paper (PDF)
The fastest way to get the application running is with Docker and Docker Compose.
-
Clone the repository:
git clone git remote add origin [email protected]:Adarsh-Roy/cd_prediction.git cd cd_prediction
-
Build and run the container:
docker build -t cd-prediction-app . && docker run -p 8501:8501 cd-prediction-app
-
Open the application:
Once the container is running, open your web browser and navigate to: http://localhost:8501
- Upload a file: Use the file uploader to select a point cloud file. The application supports both
.pcdand.paddle_tensorformats. - View the prediction: The application will process the file and display the predicted drag coefficient (Cd) value.
This repository is a monorepo containing both the Streamlit application and the complete code for model training and experimentation.
If you are interested in the details of the data preprocessing, model architecture, or training process, please refer to the detailed documentation in the training directory:
The model file present in this repository is obtained by training on DriveAer++ dataset with around 8000 car point clouds.
- Support points cloud uploads in any orientation. The current implementation only handles point clouds in standard orientation, that is, the rear end to the front end of the car is along the positive x axis.
- Support standard cad model file formats, like .stl.