About train.py !!

[Z-Z-J]

wild videos

python train.py -n wild -d 1 --kernel_size 5,3,1 --stride 3,1,1 --dilation 1,1,1 --channel 1024 --confidence 1 --translation 1 --contact 1 --loss_term 1101

First，I noticed that config.trainer.data always is "gt" 。This means only use “./data/data_h36m.npz " to train 。 Second, Can you provide './data/data_2d_h36m_cpn_ft_h36m_dbb.npz' and './data/data_2d_h36m_detectron_ft_h36m.npz' ??

[Shimingyi]

I will suggest to train our model on gt settings, because we simulate the noise distribution which is from wild video, it can make our noises more realistic. If you train it on fine-tuned detecton or cpn, the noise is just from human3.6m. And also, the training with gt will get best performance, so we only realease the these pre-train models and configration based on gt-training.

If you still want to use other dataset or 2d detection to train the model, you can refer VideoPose to get their data, and then put them into ./data folder, run command with --data cpn or --data detectron. The hyper parameters and model structure need to be updated to get similar performance(still worse than gt).

[Z-Z-J]

Thank you for such a quick reply !! Can you add me to QQ ? I am a student of Beijing University of Posts and Telecommunications ! My English is relatively poor, so I hope I can communicate with you in Chinese.After you add my QQ, I will send emails to ask you questions, without disturbing your normal life. My QQ : 1546375589 。If it is inconvenient, I will continue to ask you questions on github。No matter what, thank you!!!

[snitchjinx]

Hello Mingyi! Shall we also add QQ friends? My QQ is 908982572. I have a few questions about the project and really would like to discuss them with you. Thank you!