Jeofu
Jeofu
> @TrinhQuocNguyen Thanks for you response. @AterLuna In addition, the format of file lists are attached. You can either use absolute path or relative path for training. > > ```...
It's OK, the mask is auto generated in the code,thanks > > @TrinhQuocNguyen Thanks for you response. @AterLuna In addition, the format of file lists are attached. You can either...
sorry, my fault, I used the defalut DataLoader in my code. replaced with SafeDataLoader and solved my problem. but I have another question, the performance seems to be not very...
you should set the args `--gpu-ids = '0,1,2' ` in` train.py` and ensure the CUDA_DEVICES_VISIBLE=0,1,2 in tran.sh to ensure the multi-gpu training
hope for a feasible solution
sorry, I made a mistake about the reason. I find that the results without limb poses is irrelevant with the `body_pose_prior.pkl` after checking and digging into the code carefully. I...
the length of `pose_body_ids` of smpl is 69 according to the code in` chmosh.py`, which is different from smplx 63 ``` if cfg.surface_model.type == 'smpl': pose_body_ids = all_pose_ids[3:] #69 elif...
the problem is at the `pose_body_prior.pkl` used in` moshpp.prior.gmm_prior_ch.py`, where the `gmm['covars']` and `gmm['means']` shapes are `(8, 63, 63) `and `(8,63)` respectively in the pkl file, so the `npose=69 `doesn't...
the blender output mesh result is like #5 , the pose information is lost with correct shape and orientation.