Table 6.

Convolutional neural network architectures for pose estimation.

Model	Highlight	Source Code (Framework)	Reference
Heatmap-based networks
CPHR [212]	Detection heatmap, regression on heatmap, cascade network	GitHub [215] (Torch)	[216]
CPMs [213]	Sequential network, natural learning objective function, belief heatmap, multiple stages and views	GitHub [217] (Caffe, Python, Matlab)	[216]
Hourglass [214]	Cascaded network, hourglass module, residual connection, heatmap	GitHub [218] (Torch)	[97,104,216], etc.
Heatmap-free networks
DeepLabCut [211]	ROI, residual network, readout layer	GitHub [219] (Python, C++)	[132]
DeepPose [210]	Holistic fashion, cascade regressor, refining regressor	GitHub [220] (Chainer)	[132]

Note: CPHR is convolutional part heatmap regression; CPMs is convolutional pose machines; and ROI is region of interest.