Depth-Estimation-on-Resource-Limited-Devices

---

Table of Contents

• Description
• Install and Dependencies
• How To Use
• Author Info

---

Description

@@ Novel Depth Estimation Model for Resource-limited Devies @@
@@ Bachelor's thesis project. HSE, Moscow. 2021 @@

Novel Depth Estimation Model for Resource-limited Devies Bachelor's thesis project. HSE, Moscow. 2021

SOTA Depth Estimation deep learning models usually require high computational resource in order to operate. Neural-Architecture-Search (NAS) is a technique for automating the designing of artificial neural networks without assistance of field expertise.

We have used NAS (ProxylessNAS) approach to design a novel depth estimation model with high latency and accuracy.

Table of Contents

• dependencies.txt : Required packages to run
• mymain.py : To start neural architecture search process
• mymodel.py : Base model of the search space
• ops.py : Operations (convolution, upconvolution, etc.)
• mynyu.py : NYUDepth dataset and dataloader
• transforms.py : Data transformation utility functions
• myutils.py : Utility functions

--

Install And Dependencies

some text

Dependencies

• Neural Network Intelligence

--

How To Use

  git clone https://github.com/tombuuz/Depthmap-Resource-Limited

Install required packages

  conda create -n myenv --file dependencies.txt

To install NNI

  pip install nni

or

  git clone https://github.com/microsoft/nni 

--

Author Info

Author: Dagvanorov Lkhagvajav

Email: [email protected]

Scientific Supervisor: lya Makarov