3D human pose estimation involves estimating human joint locations in 3D directly from 2D camera images. Theestimation model would have to estimate the depth information directly from the 2D images. We explore two methods in this paper both of which represent human pose as a heatmap. The first one follows (Newell et al. [6]) and (Martinez et al. [7]) where we predict 2D poses and then lift these 2D poses to 3D. The second approach is inspired by (Pavlakos et al. [8]) and involves learning 3D pose directly from the 2D images. We observe that while both these approaches work well, the mean of both their predictions gives us the best mean per-joint prediction error (MPJPE) score.