# PointNet with CLIP Features for Multi-View Object Classification

This repository contains code for training a multi-view object classification model using PointNet with CLIP (Contrastive Language-Image Pretraining) features. The model combines 2D and 3D features to perform classification on 3D point cloud data.

## Prerequisites

- Python 3.6 or later
- PyTorch 1.7 or later
- CUDA-capable GPU (optional but recommended for faster training)
- Other dependencies listed in `requirements.txt`

## Installation

1. Clone this repository:

   ```sh
   git clone https://github.com/yourusername/pointnet-clip-object-classification.git
   cd pointnet-clip-object-classification

2. Install the required packages using pip:

   pip install -r requirements.txt

Usage

1. Train the PointNet-CLIP model:

   python main.py --batch_size 32 --num_points 1024 --nepoch 250 --outf results

   Adjust the batch size, number of points, number of epochs, and output folder as needed.

2. Evaluate the trained model:

   python main.py --batch_size 32 --num_points 1024 --model results/3D_model_X.pth --outf results --feature_transform

   Replace X with the epoch number of the trained model.

3. View results:

   Training and testing progress, as well as accuracy, will be displayed during training and saved to a log.txt file in the output folder.

Acknowledgments

• This code builds upon the PointNet architecture and CLIP model.
• Realistic_Projection module is used for projecting point clouds to 2D images.
• The dataset is assumed to be organized in the specified structure under dataset_path.

Citation

If you use this code in your research, please consider citing:

@article{YourArticleCitation,
  title={Title of Your Article},
  author={Author Names},
  journal={Journal Name},
  year={Year},
  volume={Volume},
  number={Number},
  pages={Page Range},
  doi={DOI},
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Please replace `yourusername` in the repository URL with your actual GitHub username and adjust any paths or details according to your use case.