March 15, 2024
Recently, apaperwritten by an undergraduate studentsupervised by Prof. Bai Xiang, Dean ofSSE, was accepted by IEEE/CVF Computer Vision and Pattern Recognition Conference (CVPR)2024. CVPR isregarded as atop-levelinternational conference in computer vision. The paper istitledDynamic Adapter Meets Prompt Tuning: Parameter-Efficient Transfer Learning for Point Cloud Analysis,withZhou Xin, anundergraduate student of the School of Future Technology,asthe first author, and Prof. Bai Xiangasthe instructor and corresponding author.
Under the guidance of Prof. Bai Xiang, Zhou Xinsystematically studiedrelated knowledge of 3D vision, andfocused on the parameter-efficient transfer learning task for point cloud analysis. Afterconducting a number ofexperiments,Zhouproposed a lightweight dynamic adapter structure anddeveloped a prompt tuning strategy based on internal model features.Throughtheoptimizationonlyonproposed module,itsignificantly reduces 95% of the trainable parameters,and saves 50% of the training memory. At the same time, the optimized modulecan ensure that the model’sperformance is barely unaffected, whichis a cost-effectiveway foredgemodeldeployment.
CVPR is a Class-A conference recommended by the China Computer Federation (CCF). CVPR holds a high reputation in academia, which is hosted by IEEE. It is recognized as one of the top three conferences in computer vision, alongside the International Conference on Computer Vision (ICCV) and the European Conference on Computer Vision (ECCV).