This document describes a framework for 2D pose estimation using active shape models and learned entropy field approximations. A dataset of manually annotated poses was created from NBA footage to train the models. Active shape models use principal component analysis to represent poses as a linear combination of modes of variation learned from the training data. To evaluate pose likelihood, image entropy is proposed as a texture similarity measure and regression is used to learn a function mapping poses to entropy fields, which can be compared to the image entropy. Current results are presented and future work to improve and speed up the approach is discussed.