Improving Efficiency of RetinaFace Models by Using Controlled Branch Pruning
Daniel Kaempf
Miljan Vuletić
What is the research paper about?
The goal of this paper is to improve the efficiency of a RetinaFace model for face recognition in a specific situation where the detection of faces that are large in relation to the image resolution is required. The improvement is made in a way that will not adversely affect the performance of the model and does not require the model retraining. The method that will be considered is pruning the model used for detection, which means removing the branches of the model that contribute to the detection of smaller faces (in relation to the image resolution).
Why Download the Research Paper?
Enhanced Performance: Learn how controlled branch pruning improves RetinaFace models efficiency without compromising detection accuracy.
Edge AI Optimization: Discover how the pruned model is optimized for embedded and mobile devices, offering reduced memory usage and faster execution.
No Retraining Required: Gain insights into how to optimize your models without the need for complex retraining processes.