Human pose estimation from a single depth image.
Currently implements random tree walks (RTW) on the CAD-60 dataset.
(1) Process the CAD-60 dataset from data/datasets/ into numpy arrays in data/processed/.
python data/process_cad_60.py
(2) Train network using random tree walks (RTW).
python models/random-tree-walks/rtw.py
(3) View the visualizations in output/random-tree-walks/png/
depth-pose-estimation/
data/
process_cad_60.py
datasets/ # raw dataset files
CAD-60/
...
processed/ # processed numpy files
CAD-60/
depth_images.npy # depth images (N x H x W)
joints.npy # joint coordinates (N x NUM_JOINTS x 3)
...
models/
random-tree-walks/
rtw.py
helper.py
...
output/
random-tree-walks/
models/ # saved models
preds/ # saved prediction
png/ # visualizations