Pose Estimation of Multiple Domains Based on the Fusion of Multiple Deep Learning Models and Baidu API

To enhance the precision of pose estimation across various domains, this study introduces a method that leverages the general object recognition capabilities of the Baidu API. For an unfamiliar image, we first employ the output of the Baidu API to infer its domain, then proceed with keypoint detection using a pose estimation model pre-trained on relevant domain datasets. This approach has been validated through extensive experiments on three widely-used datasets: MPII, Animalpose, and Animalkingdom. In comparison to existing algorithms, our method reduces the number of high-performance models to be trained, thus minimizing computational demands and ensuring high scalability. Experimental findings confirm that our method is not only simpler but also more accurate.