AIDRImageFiltering

Image Crawler Module

Crawl images from json file extracted from twitter.

How to run:

cd AIDRImageFiltering/
java -classpath /target/image-filtering-0.0.1-SNAPSHOT.jar hbku.qcri.sc.aidr.imagecrawler.ImageCrawler -j sample/150311152434_cyclone_pam-15_20150326_vol-4.json -f sample/images -m p

Options to run:

• -j Json file of the collection
• -f Save image to the folder under the name of collection
• -m Run in parallel, value is p for parallel running option, it will check the number of processor the machine has and will utilizes 75% of it.

How to train a binary classification model (AlexNet)

• Copy trainning images to NEG and POS folder

cd AIDRImageFiltering/
java -classpath /target/image-filtering-0.0.1-SNAPSHOT.jar hbku.qcri.sc.aidr.filtering.ImageFilteringModel -d toTrain/ -n 100 -b 60 -i 2 -s savedModel

Options to run:

• -d dataset folder
• -n total number of training images
• -b batch_size
• -i interation
• -l learing rate
• -s save trained model to a folder