The COREL Database: The database is divided into 7 archives, I downloaded all of them, and merged them with the command: cat CorelDB.7z.00* > CorelDB.7z. You can find this file in ./data/CorelDB.7z.
Install the dependencies:
make install
Random split coreldb to 3 dataset (test, train, validation):
make split-dataset
Feature list: [color, daisy, edge, gabor, hog, vgg, res]
# Launch classification using color feature
make cbir -- -a color
# Launch feature fusion between daisy and color (with weight 2 and 5)
make cbir -- -f daisy:2 color:5If you do not want to use make:
# Launch classification using color feature
cd src/CBIR && python -m pipenv run python scripts/classify.py -a color
# Launch feature fusion between daisy and color (with weight 2 and 5)
cd src/CBIR && python -m pipenv run python scripts/classify.py -f daisy:2 color:5Full usage help (from make cbir -- --help):
usage: classify.py [-h] [-D DEPTH] (-a {color,daisy,edge,gabor,hog,vgg,res} | -f FUSION [FUSION ...])
optional arguments:
-h, --help show this help message and exit
-n NEIGHBOR, --neighbor NEIGHBOR
neighbor by class
-a {color,daisy,edge,gabor,hog,vgg,res}, --feature {color,daisy,edge,gabor,hog,vgg,res}
Feature to launch
-f FUSION [FUSION ...], --fusion FUSION [FUSION ...]
Use feature fusion method
# Launch CNN classification with default args
make cnn
# Launch CNN classification with epochs=50 and rewrite weights (re-train the model)
make cnn -- -e 50 -RIf you do not want to use make:
# Launch CNN classification with default args
cd src/CNN && python -m pipenv run python neural_model.py
# Launch CNN classification with epochs=50 and rewrite weights (re-train the model)
cd src/CNN && python -m pipenv run python neural_model.py -e 50 -RFull usage help (from make cnn -- --help):
usage: neural_model.py [-h] [-e EPOCHS] [-b BATCH_SIZE] [-R] [-H]
optional arguments:
-h, --help show this help message and exit
-e EPOCHS, --epochs EPOCHS
-b BATCH_SIZE, --batch-size BATCH_SIZE
-R, --rewrite-weights
Re-write weights
-H, --hide-plots Hide plots