Open andiconda opened 5 years ago
I have the same question. I have videos with upper body only. I would like to know how well it works if only the upper body is visible before installing. Thanks
The pre-trained models can only do full body pose estimation I believe. Thus, I trained a model to do pose estimation just for upper body by using only the relevant upper body keypoints in training to do pose estimation for upper body joints. Specifically, for the pre-trained Detectron results that are provided, I used 11 out of 17 COCO keypoints (excluding the legs) to estimate 10 out of 17 H3.6M joint positions (joints above waist). The results are better compared to the pre-trained models when the lower body is not visible in the video.
@ahmetnarman Could you explain me how you did it? I am facing a related problem. I am also interested to train the model only for the upper body.
this can be used for upper body?