sauradip / CNN_SLAM

CNN SLAM implementation of https://arxiv.org/abs/1704.03489

README

Updated: 4th October 2018.

Computer Vision and Intelligence Group, IIT Madras

---

Blog: iitmcvg.github.io

This repository contains content that we use for CNN SLAM. Original paper

---

Contributors

• Varun Sundar
• Aditya S
• Aaaryaan

---

Table of Contents

1. Download Monodepth Checkpoints
2. Using Monodepth Infer
3. Install g2o

---

Experiments

1. Heterogenous graphs; convert every CPU op into a compute graph and note performance with placement.
2. Stereo matching methods and GPU optimisation.
3. CPU efficient object detectors and depth estimators.
4. Indoor optimised depth estimation (monocular.)

---

To Do

• Complete camera pose estimation.
• Monocular depth methods library.
• Freeze graphs for inference.
• Sample runs.
• Global Graph Optimisation.

---

G3docs

• Monodepth

---

Monodepth checkpoints

---

References

1. Python Bindings for LSD SLAM
2. LSD SLAM
3. Python Bindings for ORB SLAM 2
4. ORB SLAM
5. DSO SLAM
6. A SLAM Primer