This project investigated whether supplementing a partial depth map with a small amount of user input could allow a convolutional neural network to more accurately estimate the full shape of a 3D object. The file train-model.py contains the training and testing methods and restore-model.py contains the code for rebooting the training process (in case one wants to run more iterations after it terminates normally). Our proposed user-input model was able to slightly outperform the baseline model, suggesting that there does exist a potential for user input to be helpful for the task of 3D shape estimation. The file written_final_report.pdf contains the writeup submitted as part of my indepedent work under Prof. Jia Deng during spring 2019.