3D Human Pose Estimation via Deep Neural Network

Methodology

The prediction model adapts the VGG16 architecture. Based on the pre-trained model that was used by the VGG team in the ILSVRC-2014 competition, we are using the pre-trained weight for all the convolutional layers to extract deep features from the images and fine tuning the last two dense layers on the FLIC dataset.

Dependency

Keras:Deep Learning library for Theano and TensorFlow

Sample Output

Some sample outputs on the FLIC dataset. Skeleton in green is groundtruth, while the red one is prediction from the model.

Future Work

We are actively working on accomodating the model to perform 3D pose estimation.