HPI Seminar on Semantic Multimedia, summer semester 2013
We are group 1 – Mandy Roick, Claudia Exeler, Tino Junge, and Nicolas Fricke
Our objective is to analyze Images on Flickr and build homogenous clusters of the results. Read more: http://semmul2013.blogspot.de/2013/04/seminar-challenge.html [german]
$ sudo apt-get install git # only if you do not have git already
$ git clone https://github.com/nicolas-fricke/semmul2013-group1.git$ sudo apt-get install python-pip
$ sudo pip install flask
$ sudo pip install flask_assets
$ sudo pip install nltk
$ sudo pip install SimpleCV
$ sudo pip install Pycluster
$ sudo apt-get install python-scipy python-pygame python-opencv mclopen a python console
$ pythonand use nltk to download an offline instance of wordnet
>>> import nltk
>>> nltk.download()
# opens download tool
# enter d for download
# enter wordnet for downloading wordnet
# enter q for quit$ cd semmul2013-group1/
$ cp config.cfg.template config.cfg
$ vim config.cfg
# adapt config:
# 1. data-dir: should be the path to the folder, where metadata and images can be found, for example .../semmul2013-group1/data/
# 2. metadata-dir: should be the path to the folder, where metadata jsons for Flickr-images can be found, for example .../semmul2013-group1/data/metadata
# -> folder structure should be: metadata/10-19/10/100210.json
# 3. downloaded-images-dir: if Flickr-images are already downloaded, this is the path to the folder, where they can be found
# 4. preprocessed-data-dir: should be the path to the folder, where preprocessed data can be stored, for example .../semmul2013-group1/data/preprocessed_data
# 5. keywords-for-pictures-dir: should be the path to the folder, where the results of Synset detection can be stored, for example .../semmul2013-group1/data/preprocessed_data/keywords_for_pictures
# -> all folders have to exist before usage$ cd src/
$ export PYTHONPATH=$PYTHONPATH:`pwd`
$ python clustering/semantic/preprocess_synset_detection_bestfirstsearch.py # can be processed in parallel by using parameter -d
$ python clustering/semantic/preprocess_keyword_clusters.py
$ python clustering/visual/preprocess_visual_features.py # can be processed in parallel by using parameter -d
$ python frontend/frontend.py
# now start a browser (preferably google chrome) and browse to localhost:5000