To enhance the clarity and standardization of action demonstration resources in physical education teaching, this paper has constructed a high-fidelity sports teaching demonstration video synthesis model based on generative adversarial networks, and tested its impact on the level of action cognition from the perspective of teaching application. The experimental results show that the total score of action cognition in the experimental group increased from 72.1 points to 86.9 points, while that in the control group increased from 71.8 points to 78.6 points; the improvement rates of the experimental group in action sequence recognition and key point memory were 20.3% and 19.4% respectively, both higher than those of the control group. In the delayed test, the scores of the experimental group decreased from 86.9 points to 84.4 points, while those of the control group decreased from 78.6 points to 75.5 points. The retention effect was more stable, indicating that the high-fidelity synthesized demonstration videos can effectively promote students’ understanding and memory of actions.
Povzetek: This paper takes the insufficient quality of action demonstration resources in physical education classes as the starting point, and introduces a generative adversarial network to build a high-fidelity sports teaching demonstration video synthesis model. Through methods such as key point extraction, feature representation, and temporal constraints, the standardization and coherence of the demonstration videos are improved. Combined with teaching experiments, the synthesized videos were verified in terms of the improvement of action cognitive level and learning retention effect. The results show that such demonstration videos help students better grasp the structure and key points of the actions, and have promotional value in physical education practice.
