4-neuman运行和训练bike.md

neuman运行和训练bike

输出到demo目录下

一. 运行

C:\Program Files\OpenSSL-Win64\bin 放到path
启动anaconda prompt，不是powershell
conda activate neuman_env

---

渲染人的经典动作，白背景

python render_360.py --scene_dir ./data/bike --weights_path ./out/bike_human/checkpoint.pth.tar --mode canonical_360 --use_cuda true

渲染人的姿态，在训练集里面的动作，白背景

python render_360.py --scene_dir ./data/bike --weights_path ./out/bike_human/checkpoint.pth.tar --mode posed_360 --use_cuda true

---

python render_test_views.py --scene_dir ./data/bike --weights_path ./out/bike_human/checkpoint.pth.tar --use_cuda true

带背景的，带新动作的渲染

python render_reposing.py --scene_dir ./data/bike --weights_path ./out/bike_human/checkpoint.pth.tar --motion_name=jumpandroll --use_cuda true

把parkinglot seattle citron三个场景的人渲染到seattle场景

python render_gathering.py --actors parkinglot seattle citron --scene_dir ./data/seattle --weights_path ./out/seattle_human/checkpoint.pth.tar --use_cuda true

二. 训练

训练自行车场景 python train.py --scene_dir ./data/bike/ --name=bike_background_yangkang --train_mode=bkg --use_cuda true

训练自行车场景的人 python train.py --scene_dir ./data/bike --name=bike_human_yangkang --load_background=bike_background_yangkang --train_mode=smpl_and_offset --use_cuda true

三. 训练参数

optional arguments:
  -h, --help            show this help message and exit
  --train_mode {bkg,smpl_only,smpl_and_offset}

General:
  --use_cuda USE_CUDA   cuda

NeRF:
  --nerf_depth NERF_DEPTH
                        network depth
  --nerf_width NERF_WIDTH
                        network width
  --use_viewdirs USE_VIEWDIRS
                        use directions as network input
  --specular_can SPECULAR_CAN
                        no specular in canonical space

Positional Encoding:
  --raw_pos_dim RAW_POS_DIM
                        dimension of postion(XYZ, or XY)
  --pos_min_freq POS_MIN_FREQ
  --pos_max_freq POS_MAX_FREQ
  --pos_N_freqs POS_N_FREQS
  --raw_dir_dim RAW_DIR_DIM
                        dimension of direction
  --dir_max_freq DIR_MAX_FREQ
  --dir_N_freqs DIR_N_FREQS
  --log_sampling LOG_SAMPLING
  --include_input INCLUDE_INPUT
  --can_posenc CAN_POSENC
                        rotate positional encoding for can space