voxelnerf command 짜투리

초기실험들.

 

Neuralbody command

python train_net.py --cfg_file configs/zju_mocap_exp/latent_xyzc_multi.yaml exp_name xyzc_multi_debug resume False

python run.py --type evaluate --cfg_file configs/zju_mocap_exp/latent_xyzc_multi.yaml exp_name xyzc_multi_debug eval True

IBRNet Command

python -m torch.distributed.launch --nproc_per_node=1 train.py --config configs/h36m.txt --num_source_views 3 --expname debug

python train.py --config configs/h36m.txt --num_source_views 3 --expname h36m1 

ssh -L 16006:127.0.0.1:6006 nipa

tensorboard --logdir h36m4 --samples_per_plugin images=100

PixelNeRF Command

train:

python train/train.py -n h36m_affine13 -c conf/exp/h36m.conf -D data/Human36M -V 3 --gpu_id=0 --lr 0.000003

lr은 저게 젤 잘된다.

eval:

 python eval/eval_h36m.py -c conf/exp/h36m.conf -D data/Human36M --split test -n h36m_affine13 --gpu_id=0

do_vis=True해놓으면 ${ROOT}/eval_out/{h36m_affine13/에 저장됨