Seasons Of Code

3D reconstruction using 2D images    • Siddhesh Pawar    • Hrushikesh Bodas   

3D reconstruction using 2D images

3D reconstruction using 2D images


Producing 2D images of a 3D world is inherently a lossy process, i.e. the entire geometric richness of 3D gets projected onto a single flat 2D image. We aim to create an API in Python which primarily reconstructs 3D volumes from 2D X-Ray Images.

We see this project as the first step towards a diagnostic tool in conditions where either no CT equipment or the education to interpret x-ray imagery is available, such as for mobile x-ray devices, lay users, or medical diagnostics in developing countries. The project is primarily divided into 2 parts:

  • Implementation of various CNN architectures for 3D reconstruction from 2D images(3 people would be working on this part)

  • Development of API(back-end framework for the above task).1 mentee would be working on this part.

Part 1 has some hard pre-requisites while anyone who has an interest in python or has done some basic programming in python or java-script can apply for part 2.

Pre-requisites for part 1: Must be familiar with any one of the following deep learning frameworks: Pytorch/Tensorflow/Theano/Keras. A basic idea of neural networks and machine learning is required. Previous experience in image processing is desired although is not a hard pre-requisite.

Interested people in this part should go through the following paper while applying:

Note: If you are new to deep learning, it is recommended that you should go through the first 5 chapters of the book before applying:

Tentative Timeline:

Week Work
Week 1 and 2(Pre-Endsems) Reading of related material and learning relevant applications of the framework that would be used(mostly Keras and PyTorch)
Week 3 Testing and implementing Simple CNN architectures
Week 4 and 5 Working on Designing and implementation of 3D reconstruction from multiple images along with data pre-processing
Week 6 and 7 Programming and testing of various models for 3D reconstruction from single 2D image
Week 8 Further improvements on the models that have been created above.