https://github.com/ugnelis/find-path

Find path by using Semantic Segmentation.
https://github.com/ugnelis/find-path

dataset dataset-maker fcn-16s ipynb semantic-segmentation tensorflow vgg16

Last synced: 6 days ago
JSON representation

Find path by using Semantic Segmentation.

• Host: GitHub
• URL: https://github.com/ugnelis/find-path
• Owner: ugnelis
• License: mit
• Archived: true
• Created: 2016-11-11T15:46:06.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2017-09-26T17:21:48.000Z (about 7 years ago)
• Last Synced: 2024-09-23T22:02:06.506Z (10 days ago)
• Topics: dataset, dataset-maker, fcn-16s, ipynb, semantic-segmentation, tensorflow, vgg16
• Language: Jupyter Notebook
• Homepage:
• Size: 4.03 MB
• Stars: 10
• Watchers: 1
• Forks: 5
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# find-path

Find Path project finds humans paths and routes, such as sidewalks, park ways, forest paths. This project implements semantic segmentation approach. It uses VGG16 pretrained model.

[![Semantic Segmentation - Ugnius Malūkas](https://img.youtube.com/vi/jTj_mzeoVm0/0.jpg)](https://www.youtube.com/watch?v=jTj_mzeoVm0)

## Path Finding 

### Installation 

TODO

### Training

Go to *calculations* folder.

```bash

$ cd calculations

```

Run training.

```bash

$ python train.py

```

### Run Trained Model

Check for examples in *calculations/demo.py*, *calculations/demo.ipynb*, *calclulations/video_demo.ipynb* files.

## Dataset Maker

For making dataset, web based application was made which uses just JavaScript without any framework.

### Installation

Before installation, make sure that NodeJS, npm and bower are installed.

```bash

$ cd dataset_maker

$ bower install

```

### Launching

Open *index.html* and have fun.

## Dataset

All dataset images have 320 width, 180 height and contain 3 channels. Every image has own *.json* file which describes object in the image. In this project only 3 classes are observed: **boundaries** (everything arround path), **paths / ways** and **obstacles** (things that are on path - eg. human, road pit and etc.). Dataset contains 300 images (I'll put a bit later).

## Path Finding with OpenCV

Check *calculations/cv/* folder.

## Code References

https://github.com/MarvinTeichmann/tensorflow-fcn

https://github.com/shelhamer/fcn.berkeleyvision.org

https://github.com/machrisaa/tensorflow-vgg