Human 3D Pose Estimation Based on Inception Architecture

Researchers provide us with newer and deeper structures of neural networks every year, confirming their effectiveness concerning earlier versions of their architectures. A common assumption in creating current structures is to ignore performance issues. On the one hand, we theoretically obtain better and better classification and regression methods. On the other hand, despite the significant development of mobile devices, we lose the possibility of their implementation in systems accessible to humans, such as mobile phones. Being aware of social expectations towards the technologies being developed, we should create algorithms that improve our previous versions’ operation and optimize performance. The study compares the operation of Inception-V3 and V4 networks in precision and speed in the regression process. The estimation ability was determined as part of studying the position of human joints in 3D space. The LoCO architecture, one of the leading 3D human pose estimation methods, was used for the experiments.