Answers are grounded in this paper and cite their sources.
The problem of lagging action feedback and relying on subjective judgment in physical education classes in primary and secondary schools has long existed. In this paper, a lightweight spatio-temporal convolution model based on improved structure of Ghost Network is constructed. By introducing an improved Ghost module in the spatial dimension to generate redundant features by linear operation, the calculation amount is reduced. The dynamic information of continuous actions is extracted by depthwise separable temporary convolution, and the skeleton structure and convolution representation are supervised by the thermal map of keypoints of human body, and the action quality judgment and error location are realized by combining the differential error correction vector field based on angle deviation. This study provides a practical path for the real-time performance, evaluation accuracy, and error correction interpretation of sports actions.
No takes yet. Share an insight, caveat, or question.
Chunhe Deng (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: