Welcome to the Multi-Modal-Cuda-C-GAN repository! This project features a cutting-edge 3D deep learning model implemented in CUDA/C, specifically a Hybrid GAN that integrates cross-attention, self-attention, and convolutional blocks within the generator. The model leverages C for high-performance and scalable deep learning solutions.
- Self-Attention: Enhances the generator's capability, implemented in C for flexibility.
- Cross-Attention Mechanism: Designed for 3D tensors, suitable for CNN layers. It merges separate input tensors, outputting the same size, facilitating multi-input images and new data introduction during processing. Available in C and C++.
- Convolutional Blocks: Core convolution operations for the GAN, implemented in C for efficiency.
- GAN Structure: Comprehensive GAN architecture featuring a UNet within the generator, implemented in C for robust performance.
We are actively developing the training script in C, recreating many dependencies typically found in PyTorch from scratch, to ensure optimal performance and customization.
- Implement GAN training and code iteration in C and C++.
- Update the README with detailed setup instructions and usage examples.
- Optimize nested for loops and arithmetic operations for memory efficiency.
- C Compiler: GCC or equivalent
-
Clone the repository:
git clone https://github.com/your-username/Multi-Modal-Cuda-C-GAN.git cd Multi-Modal-Cuda-C-GAN -
Compile the code:
gcc -o main main.c -lcuda
We welcome contributions! Please check the issues for tasks that need assistance or open a new issue to propose enhancements and report bugs.
This project is licensed under the MIT License. See the LICENSE file for details.
For any questions or feedback, please contact aha.andresen@gmail.com.