This is the Tensorflow implementation of Convolutional Pose Machines, one of the state-of-the-art models for 2D body and hand pose estimation.

• Easy multi-stage graph construction
• Kalman filters for smooth pose estimation

• Windows 10 / Ubuntu 16.04
• Tensorflow 1.2.0
• OpenCV 3.2

Put downloaded models in the models/weights folder.

• Body Pose Model
• Hand Pose Model

There are two scripts, demo_cpm_body.py for body pose estimation and demo_cpm_hand.py for hand pose estimation. I take demo_cpm_hand.py for example.

First set the DEMO_TYPE. If you want to pass an image, then put the path to image here. If you want a live demo through a webcam, there are few options.

• MULTI will show multiple stages output heatmaps and the final pose estimation simultaneously.
• SINGLE will only show the final pose estimation.
• HM will show each joint heatmap of last stage separately.

The CPM structure assumes the body or hand you want to estimate is located in the middle of the frame. If you want to avoid that, one way is to add a detector at the begining, and feed the detected bounding box image into this model.

See models/nets for model definition, I take models/nets/cpm_hand.py for example.

• Create a model instance
• Set how many stages you want the model to have (at least 2)
• Call build_loss if you want to do the training
• Use self.train_op to optimize the model

Please see train.py for an example.