VIBE
VIBE copied to clipboard
mkocabas
what modification do I need if I want to train using Human3.6 dataset
Hello Thank you for the excellent work I wander to know the differences between the "J_regressor_extra" and "J_regressor_h36m" Does this difference will affect the performance or correctness of the training procedure. I ask this question because I encounter some problems when I train the model using human36 dataset(I preprocess the dataset according to the "threedpw_util.py" code and do some modification) but my training results are abnormal： the SMPL parameter was ok but camera parameter was wrong，just like [0.3, -8.1, -6.5], i think the abs() of last two numbers can't be such large, this camera will lead to wrong 2d projection. And during the training procedure, the abs() of last two numbers become larger and always minus. but the 2d keypoint loss finally decrease. i check many days but don't found the mistakes. so does the "J_regressor" will lead to this mistake by using human36 dataset or I missed something else? I also hope you can give me some advices about training using human36 dataset or could you provide the code you process the human36 dataset. Hope your reply, thanks!
Hello Thank you for the excellent work I wander to know the differences between the "J_regressor_extra" and "J_regressor_h36m" Does this difference will affect the performance or correctness of the training procedure. I ask this question because I encounter some problems when I train the model using human36 dataset(I preprocess the dataset according to the "threedpw_util.py" code and do some modification) but my training results are abnormal： the SMPL parameter was ok but camera parameter was wrong，just like [0.3, -8.1, -6.5], i think the abs() of last two numbers can't be such large, this camera will lead to wrong 2d projection. And during the training procedure, the abs() of last two numbers become larger and always minus. but the 2d keypoint loss finally decrease. i check many days but don't found the mistakes. so does the "J_regressor" will lead to this mistake by using human36 dataset or I missed something else? I also hope you can give me some advices about training using human36 dataset or could you provide the code you process the human36 dataset. Hope your reply, thanks!
Hello. Could you please release your codes for preprocessing the h36m dataset? It will be appreciated! Thank you!
Hello Thank you for the excellent work I wander to know the differences between the "J_regressor_extra" and "J_regressor_h36m" Does this difference will affect the performance or correctness of the training procedure. I ask this question because I encounter some problems when I train the model using human36 dataset(I preprocess the dataset according to the "threedpw_util.py" code and do some modification) but my training results are abnormal： the SMPL parameter was ok but camera parameter was wrong，just like [0.3, -8.1, -6.5], i think the abs() of last two numbers can't be such large, this camera will lead to wrong 2d projection. And during the training procedure, the abs() of last two numbers become larger and always minus. but the 2d keypoint loss finally decrease. i check many days but don't found the mistakes. so does the "J_regressor" will lead to this mistake by using human36 dataset or I missed something else? I also hope you can give me some advices about training using human36 dataset or could you provide the code you process the human36 dataset. Hope your reply, thanks!
Hello. Could you please release your codes for preprocessing the h36m dataset? It will be appreciated! Thank you!
Hello. Have you got the codes for preprocessing the h36m and Could you please release it? Thank you!
