Training Samples Set Simplified Method based on Key Frame Extraction
Jiguang Mo, Yang Yang, Qiyue Sun, Yunxia Liu, Xuesong Gao · 2021
The requirement of huge amount of training set data has restricted the development of the deep learning his paper designs a system which uses the differential key frame extraction method to extract the key images from the original video sequences. And these key images will be used as the training set of the deep learning model. The system can simplify the training set while ensuring the recognition accuracy of the model. The experimental results show that the proposed method can reduce the manually annotate without accuracy loss as well as improving the training efficiency.