This pull request introduces several enhancements and new features to the Lite AI Toolkit, particularly focusing on the TensorRT-based face fusion and image processing pipelines. The most significant changes include the addition of a new face fusion pipeline, improvements to the Stable Diffusion pipeline, and updates to the configuration and utility files.

New Features and Enhancements:

• Face Fusion Pipeline:

  • Added a new face fusion pipeline for GPU in trt_facefusion_pipeline_gpu.h and trt_facefusion_pipeline_gpu.cpp, including methods for model initialization, detection, and landmark processing. [1] [2]

• Stable Diffusion Pipeline:

  • Enhanced the Stable Diffusion pipeline by adding support for VAE encoding and improved image preprocessing functions in trt_pipeline.cpp and trt_pipeline.h. [1] [2] [3] [4]
  • Improved the UNet inference process with additional steps for handling timesteps and noise prediction in trt_unet.cpp and trt_unet.h. [1] [2] [3]

Configuration and Utility Updates:

• Configuration Files:

  • Updated trt_config.h to define pipeline modes using an enum class for better type safety.

• Header Inclusions:

  • Added necessary header inclusions in lite.h to ensure all configurations are properly included.

These changes collectively enhance the functionality and robustness of the Lite AI Toolkit, particularly in the areas of face fusion and image processing using TensorRT.