Author: Nirshal Chandra Sekar

Profile: https://nirshalchandrasekar.github.io/

This project was a part of the course CSCI 5980: Deep Learning for Robot Manipulation at the University of Minnesota, where the objective was to gain experience building and training convolutional neural networks for 6 degrees-of-freedom rigid body pose estimation.

In this project I implemented a version of PoseCNN for object pose estimation.

The goals of this project are as follows:

• Implement the PoseCNN architecture for object pose estimation.
• Understand the characteristics of neural network based object pose estimation using the PROPS Pose Dataset.
• Gain experience reimplementing network architectures by translating from text and figure descriptions to code implementations.

PoseCNN

The network is split into three branches as follows:

• The segmentation branch
• The 3D translation branch
• The 3D rotation branch

Segmentation Branch