cqtan / imi-unicorns Goto Github PK

View Code? Open in Web Editor NEW

5.0 5.0 1.0 452.68 MB

HTW IMI-Project: Chasing Unicorns and Vampires in a Library ---> Check the Wiki

Python 70.27% Ruby 0.09% JavaScript 0.07% TypeScript 9.65% HTML 1.53% CSS 18.38%

imi-unicorns's People

Contributors

Stargazers

Watchers

Forkers

elektrobohemian

imi-unicorns's Issues

Showtime Projektbeschreibung

Script: DataProcessing

Errors beseitigen und Notebook in ein vernnftiges Script umwandeln. Evtl. auf verschiedene Funktionen aufteilen

Test: Category Feature

Test feature to filter images by their content:

Drop Down GUI to select one of the following:
- select person
- select maps
- select text
- etc.

Download images

Decide on Framework

Angular, React, or something else completely.

Showtime: Test-run our WebApp for showtime!

Make a stable version of our WebApp and perform a test run!

Showtime: T-Shirts

Decide on T-shirt design and order them

ML: JSON structure for image feature

# Test with 10 images:

new Collection for each Image
{
  features: <List>
  path: <string> -> ("dist/ChasingUnicornsAndVampires/assets/images/" + PPN + <imageName.ending>) 
  PPN: <string>
}

Insert for every Book
{
 "PPN-Nr": <List of Features>
 "PPN-Nr": ...
  ...
}

For Category
{
  feature: <Collection>(Category: featureName)
}

ML: Clustering

Research Clustering Methods

Feature extraction with pre-trained CNN

ML: by Location

Classify images according to where they were first published.
Currently using heatmaps to display location

Overview could be impoved when too many books are in same place
Try displaying as list?

Wrapping-Up: Create documentation

Create documentation in a way that other uses are able to run the app locally

Try making a wiki page for easy collaboration.
Reduce required libs as much as possible
Include requirements.txt and/or package.json file
Add version of used technologies (only the main ones, e.g. Angular, Keras, etc.)

Backend: Setup DB and File server

Currently using MongoDB and Flask.

Test: Real SBB images on our Webapp

Upload the image data to our MongoDB and display images referenced by them on the Webapp.

Extend dominant color script

Calculate most different and frequent colors

ML: Scikit Image & Feature Extraction

Research best/easiest way to extract features

Download needed parts of mets mods

Just mockups for the wiki

ML: Filter out less-meaningful images

Images, such as covers, blank pages, etc. should be clustered and filtered out so that mostly meaningful images can be displayed and to reduce the number of images in total to handle.

ML: Object Recognition for Category Feature

MIttels Object Detection bestimmte Objekte in den Bilder erkennen:

Person
Animal
- Dog
- Cat
- Unicorn
Building
Plant

DB: Add new data for images in the DB

Create a file format (JSON) to add more data for each image in the databank.
Possible data:

people
map
text
animal
etc...

ML: Dominant color

Classify images according to their dominant color.

ML: Bag of Visual Words -> Agglomerativ

Ich würde mich mal gerne mit diesem Ansatz hier beschäftigen. BovW

Neu
Agglomerativen Ansatz mit Bag, um Feature Vektoren zu sammlen (Bildähnlichlkeiten)

ML: Transfer Learning for Category Feature

Verwende gerade 4 pre-trained models aus Keras, die eng mit ImageNet arbeitet.
- https://keras.rstudio.com/articles/applications.html
- Models: VGG16, ResNet50, Inception v3, MobileNet
Diese models haben zumindest 1000 verschiedene Klassen, aber ich muss noch schauen, welche der 4 bessere Performance bietet.
NEU: Transfer Learning für nur eine handvolle verschiedene Klassen