Block-wise Quantization
- Mentors
- FYtao
- Organization
- OpenCV
- Technologies
- opencv, c++, onnx
- Topics
- computer vision, Model quantization, Deep Leaning
The proposal aims to expand upon the advancements made in the 2023 project, which focused on introducing 8-bit model quantization to OpenCV-zoo. Building upon this foundation, I added Block-wise quantization APIs to further enhance the capabilities of the framework.
The primary objective of this project is to develop a comprehensive command line tool within OpenCV-zoo and add inference support to blockwise quantization operations in OpenCV. This tool will facilitate block-wise quantization of ONNX models, offering users a robust and efficient method to optimize their models for deployment across various platforms.